src/HOL/Tools/Sledgehammer/sledgehammer_mash.ML
author blanchet
Thu, 06 Dec 2012 11:25:10 +0100
changeset 50397 d84a5ab736bb
parent 50396 5c9a2f5ab323
child 50398 78c7b52dbadb
permissions -rw-r--r--
reduce max number of dependencies for MaSh to get rid of junk
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
48380
d4b7c7be3116 renamed ML files
blanchet
parents: 48379
diff changeset
     1
(*  Title:      HOL/Tools/Sledgehammer/sledgehammer_mash.ML
48248
b6eb45a52c28 split relevance filter code into three files
blanchet
parents:
diff changeset
     2
    Author:     Jasmin Blanchette, TU Muenchen
b6eb45a52c28 split relevance filter code into three files
blanchet
parents:
diff changeset
     3
b6eb45a52c28 split relevance filter code into three files
blanchet
parents:
diff changeset
     4
Sledgehammer's machine-learning-based relevance filter (MaSh).
b6eb45a52c28 split relevance filter code into three files
blanchet
parents:
diff changeset
     5
*)
b6eb45a52c28 split relevance filter code into three files
blanchet
parents:
diff changeset
     6
48381
1b7d798460bb renamed ML structures
blanchet
parents: 48380
diff changeset
     7
signature SLEDGEHAMMER_MASH =
48248
b6eb45a52c28 split relevance filter code into three files
blanchet
parents:
diff changeset
     8
sig
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
     9
  type stature = ATP_Problem_Generate.stature
48296
e7f01b7e244e gracefully handle the case of empty theories when going up the accessibility chain
blanchet
parents: 48293
diff changeset
    10
  type fact = Sledgehammer_Fact.fact
e7f01b7e244e gracefully handle the case of empty theories when going up the accessibility chain
blanchet
parents: 48293
diff changeset
    11
  type fact_override = Sledgehammer_Fact.fact_override
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
    12
  type params = Sledgehammer_Provers.params
48288
255c6e1fd505 rationalize relevance filter, slowing moving code from Iter to MaSh
blanchet
parents: 48251
diff changeset
    13
  type relevance_fudge = Sledgehammer_Provers.relevance_fudge
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
    14
  type prover_result = Sledgehammer_Provers.prover_result
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
    15
48308
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
    16
  val trace : bool Config.T
48319
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
    17
  val MaShN : string
48379
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
    18
  val mepoN : string
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
    19
  val mashN : string
48314
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
    20
  val meshN : string
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
    21
  val unlearnN : string
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
    22
  val learn_isarN : string
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
    23
  val learn_atpN : string
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
    24
  val relearn_isarN : string
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
    25
  val relearn_atpN : string
48314
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
    26
  val fact_filters : string list
48303
f1d135d0ea69 improved MaSh string escaping and make more operations string-based
blanchet
parents: 48302
diff changeset
    27
  val escape_meta : string -> string
f1d135d0ea69 improved MaSh string escaping and make more operations string-based
blanchet
parents: 48302
diff changeset
    28
  val escape_metas : string list -> string
48308
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
    29
  val unescape_meta : string -> string
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
    30
  val unescape_metas : string -> string list
50356
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
    31
  val encode_features : (string * real) list -> string
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    32
  val extract_query : string -> string * (string * real) list
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
    33
  val mash_CLEAR : Proof.context -> unit
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
    34
  val mash_ADD :
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
    35
    Proof.context -> bool
50356
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
    36
    -> (string * string list * (string * real) list * string list) list -> unit
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
    37
  val mash_REPROVE :
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
    38
    Proof.context -> bool -> (string * string list) list -> unit
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
    39
  val mash_QUERY :
50356
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
    40
    Proof.context -> bool -> int -> string list * (string * real) list
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
    41
    -> (string * real) list
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
    42
  val mash_unlearn : Proof.context -> unit
48378
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
    43
  val nickname_of : thm -> string
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    44
  val suggested_facts :
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    45
    (string * 'a) list -> ('b * thm) list -> (('b * thm) * 'a) list
48321
c552d7f1720b learn from minimized ATP proofs
blanchet
parents: 48320
diff changeset
    46
  val mesh_facts :
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
    47
    int -> (real * ((('a * thm) * real) list * ('a * thm) list)) list
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
    48
    -> ('a * thm) list
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
    49
  val theory_ord : theory * theory -> order
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
    50
  val thm_ord : thm * thm -> order
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
    51
  val goal_of_thm : theory -> thm -> thm
48321
c552d7f1720b learn from minimized ATP proofs
blanchet
parents: 48320
diff changeset
    52
  val run_prover_for_mash :
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
    53
    Proof.context -> params -> string -> fact list -> thm -> prover_result
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
    54
  val features_of :
50356
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
    55
    Proof.context -> string -> theory -> stature -> term list
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
    56
    -> (string * real) list
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
    57
  val isar_dependencies_of : unit Symtab.table -> thm -> string list option
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
    58
  val atp_dependencies_of :
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
    59
    Proof.context -> params -> string -> int -> fact list -> unit Symtab.table
48665
14b0732c72f7 crank up max number of dependencies
blanchet
parents: 48656
diff changeset
    60
    -> thm -> bool * string list option
50382
blanchet
parents: 50359
diff changeset
    61
  val weight_mash_facts : ('a * thm) list -> (('a * thm) * real) list
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    62
  val mash_suggested_facts :
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
    63
    Proof.context -> params -> string -> int -> term list -> term -> fact list
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
    64
    -> fact list
48383
df75b2d7e26a learn command in MaSh
blanchet
parents: 48382
diff changeset
    65
  val mash_learn_proof :
48384
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
    66
    Proof.context -> params -> string -> term -> ('a * thm) list -> thm list
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
    67
    -> unit
48395
85a7fb65507a learning should honor the fact override and the chained facts
blanchet
parents: 48394
diff changeset
    68
  val mash_learn :
85a7fb65507a learning should honor the fact override and the chained facts
blanchet
parents: 48394
diff changeset
    69
    Proof.context -> params -> fact_override -> thm list -> bool -> unit
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
    70
  val is_mash_enabled : unit -> bool
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
    71
  val mash_can_suggest_facts : Proof.context -> bool
48288
255c6e1fd505 rationalize relevance filter, slowing moving code from Iter to MaSh
blanchet
parents: 48251
diff changeset
    72
  val relevant_facts :
48292
7fcee834c7f5 more code rationalization in relevance filter
blanchet
parents: 48289
diff changeset
    73
    Proof.context -> params -> string -> int -> fact_override -> term list
48296
e7f01b7e244e gracefully handle the case of empty theories when going up the accessibility chain
blanchet
parents: 48293
diff changeset
    74
    -> term -> fact list -> fact list
48319
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
    75
  val kill_learners : unit -> unit
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
    76
  val running_learners : unit -> unit
48248
b6eb45a52c28 split relevance filter code into three files
blanchet
parents:
diff changeset
    77
end;
b6eb45a52c28 split relevance filter code into three files
blanchet
parents:
diff changeset
    78
48381
1b7d798460bb renamed ML structures
blanchet
parents: 48380
diff changeset
    79
structure Sledgehammer_MaSh : SLEDGEHAMMER_MASH =
48248
b6eb45a52c28 split relevance filter code into three files
blanchet
parents:
diff changeset
    80
struct
48249
2bd242c56c90 dummy implementation
blanchet
parents: 48248
diff changeset
    81
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
    82
open ATP_Util
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
    83
open ATP_Problem_Generate
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
    84
open Sledgehammer_Util
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
    85
open Sledgehammer_Fact
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
    86
open Sledgehammer_Provers
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
    87
open Sledgehammer_Minimize
48381
1b7d798460bb renamed ML structures
blanchet
parents: 48380
diff changeset
    88
open Sledgehammer_MePo
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
    89
48308
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
    90
val trace =
48380
d4b7c7be3116 renamed ML files
blanchet
parents: 48379
diff changeset
    91
  Attrib.setup_config_bool @{binding sledgehammer_mash_trace} (K false)
48308
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
    92
fun trace_msg ctxt msg = if Config.get ctxt trace then tracing (msg ()) else ()
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
    93
48319
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
    94
val MaShN = "MaSh"
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
    95
48379
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
    96
val mepoN = "mepo"
48314
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
    97
val mashN = "mash"
48379
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
    98
val meshN = "mesh"
48314
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
    99
48379
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
   100
val fact_filters = [meshN, mepoN, mashN]
48314
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
   101
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   102
val unlearnN = "unlearn"
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   103
val learn_isarN = "learn_isar"
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   104
val learn_atpN = "learn_atp"
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   105
val relearn_isarN = "relearn_isar"
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   106
val relearn_atpN = "relearn_atp"
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   107
48394
82fc8c956cdc fixed various issues with MaSh's file handling + tune output + generate local facts again + handle nameless facts gracefully
blanchet
parents: 48392
diff changeset
   108
fun mash_model_dir () =
50340
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
   109
  Path.explode "$ISABELLE_HOME_USER/mash" |> tap Isabelle_System.mkdir
48394
82fc8c956cdc fixed various issues with MaSh's file handling + tune output + generate local facts again + handle nameless facts gracefully
blanchet
parents: 48392
diff changeset
   110
val mash_state_dir = mash_model_dir
50310
b00eeb8e352e proper quoting of paths in MaSh
blanchet
parents: 50229
diff changeset
   111
fun mash_state_file () = Path.append (mash_state_dir ()) (Path.explode "state")
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   112
48330
192444b53e86 speed up MaSh queries
blanchet
parents: 48329
diff changeset
   113
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   114
(*** Low-level communication with MaSh ***)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   115
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   116
fun wipe_out_file file = (try (File.rm o Path.explode) file; ())
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   117
50335
b17b05c8d4a4 proper quoting of paths in MaSh shell script, take 2 (cf. b00eeb8e352e)
blanchet
parents: 50319
diff changeset
   118
fun write_file banner (xs, f) path =
b17b05c8d4a4 proper quoting of paths in MaSh shell script, take 2 (cf. b00eeb8e352e)
blanchet
parents: 50319
diff changeset
   119
  (case banner of SOME s => File.write path s | NONE => ();
b17b05c8d4a4 proper quoting of paths in MaSh shell script, take 2 (cf. b00eeb8e352e)
blanchet
parents: 50319
diff changeset
   120
   xs |> chunk_list 500
b17b05c8d4a4 proper quoting of paths in MaSh shell script, take 2 (cf. b00eeb8e352e)
blanchet
parents: 50319
diff changeset
   121
      |> List.app (File.append path o space_implode "" o map f))
50319
dddcaeb92e11 robust writing of MaSh state -- better drop learning data than cause other problems in Sledgehammer
blanchet
parents: 50311
diff changeset
   122
  handle IO.Io _ => ()
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   123
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   124
fun run_mash_tool ctxt overlord save max_suggs write_cmds read_suggs =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   125
  let
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   126
    val (temp_dir, serial) =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   127
      if overlord then (getenv "ISABELLE_HOME_USER", "")
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   128
      else (getenv "ISABELLE_TMP", serial_string ())
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   129
    val log_file = if overlord then temp_dir ^ "/mash_log" else "/dev/null"
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   130
    val err_file = temp_dir ^ "/mash_err" ^ serial
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   131
    val sugg_file = temp_dir ^ "/mash_suggs" ^ serial
50335
b17b05c8d4a4 proper quoting of paths in MaSh shell script, take 2 (cf. b00eeb8e352e)
blanchet
parents: 50319
diff changeset
   132
    val sugg_path = Path.explode sugg_file
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   133
    val cmd_file = temp_dir ^ "/mash_commands" ^ serial
50335
b17b05c8d4a4 proper quoting of paths in MaSh shell script, take 2 (cf. b00eeb8e352e)
blanchet
parents: 50319
diff changeset
   134
    val cmd_path = Path.explode cmd_file
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   135
    val core =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   136
      "--inputFile " ^ cmd_file ^ " --predictions " ^ sugg_file ^
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   137
      " --numberOfPredictions " ^ string_of_int max_suggs ^
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   138
      (if save then " --saveModel" else "")
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   139
    val command =
50335
b17b05c8d4a4 proper quoting of paths in MaSh shell script, take 2 (cf. b00eeb8e352e)
blanchet
parents: 50319
diff changeset
   140
      "\"$ISABELLE_SLEDGEHAMMER_MASH/src/mash.py\" --quiet --outputDir " ^
b17b05c8d4a4 proper quoting of paths in MaSh shell script, take 2 (cf. b00eeb8e352e)
blanchet
parents: 50319
diff changeset
   141
      File.shell_path (mash_model_dir ()) ^ " --log " ^ log_file ^ " " ^ core ^
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   142
      " >& " ^ err_file
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   143
      |> tap (fn _ => trace_msg ctxt (fn () =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   144
             case try File.read (Path.explode err_file) of
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   145
               NONE => "Done"
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   146
             | SOME "" => "Done"
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   147
             | SOME s => "Error: " ^ elide_string 1000 s))
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   148
    fun run_on () =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   149
      (Isabelle_System.bash command;
50335
b17b05c8d4a4 proper quoting of paths in MaSh shell script, take 2 (cf. b00eeb8e352e)
blanchet
parents: 50319
diff changeset
   150
       read_suggs (fn () => try File.read_lines sugg_path |> these))
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   151
    fun clean_up () =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   152
      if overlord then ()
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   153
      else List.app wipe_out_file [err_file, sugg_file, cmd_file]
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   154
  in
50335
b17b05c8d4a4 proper quoting of paths in MaSh shell script, take 2 (cf. b00eeb8e352e)
blanchet
parents: 50319
diff changeset
   155
    write_file (SOME "") ([], K "") sugg_path;
b17b05c8d4a4 proper quoting of paths in MaSh shell script, take 2 (cf. b00eeb8e352e)
blanchet
parents: 50319
diff changeset
   156
    write_file (SOME "") write_cmds cmd_path;
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   157
    trace_msg ctxt (fn () => "Running " ^ command);
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   158
    with_cleanup clean_up run_on ()
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   159
  end
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   160
48308
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
   161
fun meta_char c =
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   162
  if Char.isAlphaNum c orelse c = #"_" orelse c = #"." orelse c = #"(" orelse
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   163
     c = #")" orelse c = #"," then
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   164
    String.str c
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   165
  else
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   166
    (* fixed width, in case more digits follow *)
48395
85a7fb65507a learning should honor the fact override and the chained facts
blanchet
parents: 48394
diff changeset
   167
    "%" ^ stringN_of_int 3 (Char.ord c)
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   168
48308
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
   169
fun unmeta_chars accum [] = String.implode (rev accum)
48395
85a7fb65507a learning should honor the fact override and the chained facts
blanchet
parents: 48394
diff changeset
   170
  | unmeta_chars accum (#"%" :: d1 :: d2 :: d3 :: cs) =
48308
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
   171
    (case Int.fromString (String.implode [d1, d2, d3]) of
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
   172
       SOME n => unmeta_chars (Char.chr n :: accum) cs
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
   173
     | NONE => "" (* error *))
48395
85a7fb65507a learning should honor the fact override and the chained facts
blanchet
parents: 48394
diff changeset
   174
  | unmeta_chars _ (#"%" :: _) = "" (* error *)
48308
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
   175
  | unmeta_chars accum (c :: cs) = unmeta_chars (c :: accum) cs
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
   176
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
   177
val escape_meta = String.translate meta_char
48303
f1d135d0ea69 improved MaSh string escaping and make more operations string-based
blanchet
parents: 48302
diff changeset
   178
val escape_metas = map escape_meta #> space_implode " "
48315
82d6e46c673f fixed order of accessibles + other tweaks to MaSh
blanchet
parents: 48314
diff changeset
   179
val unescape_meta = String.explode #> unmeta_chars []
82d6e46c673f fixed order of accessibles + other tweaks to MaSh
blanchet
parents: 48314
diff changeset
   180
val unescape_metas =
82d6e46c673f fixed order of accessibles + other tweaks to MaSh
blanchet
parents: 48314
diff changeset
   181
  space_explode " " #> filter_out (curry (op =) "") #> map unescape_meta
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   182
50356
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
   183
fun encode_feature (name, weight) =
50389
ad0ac9112d2c simplify code now that "mash.py" supports weights
blanchet
parents: 50383
diff changeset
   184
  escape_meta name ^
ad0ac9112d2c simplify code now that "mash.py" supports weights
blanchet
parents: 50383
diff changeset
   185
  (if Real.== (weight, 1.0) then "" else "=" ^ Real.toString weight)
50356
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
   186
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
   187
val encode_features = map encode_feature #> space_implode " "
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
   188
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   189
fun str_of_add (name, parents, feats, deps) =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   190
  "! " ^ escape_meta name ^ ": " ^ escape_metas parents ^ "; " ^
50356
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
   191
  encode_features feats ^ "; " ^ escape_metas deps ^ "\n"
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   192
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   193
fun str_of_reprove (name, deps) =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   194
  "p " ^ escape_meta name ^ ": " ^ escape_metas deps ^ "\n"
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   195
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   196
fun str_of_query (parents, feats) =
50356
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
   197
  "? " ^ escape_metas parents ^ "; " ^ encode_features feats ^ "\n"
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   198
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   199
fun extract_suggestion sugg =
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   200
  case space_explode "=" sugg of
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   201
    [name, weight] =>
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   202
    SOME (unescape_meta name, Real.fromString weight |> the_default 0.0)
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   203
  | _ => NONE
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   204
48311
3c4e10606567 implemented MaSh QUERY operation
blanchet
parents: 48309
diff changeset
   205
fun extract_query line =
3c4e10606567 implemented MaSh QUERY operation
blanchet
parents: 48309
diff changeset
   206
  case space_explode ":" line of
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   207
    [goal, suggs] =>
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   208
    (unescape_meta goal,
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   209
     map_filter extract_suggestion (space_explode " " suggs))
48312
b40722a81ac9 implemented meshing of Iter and MaSh results
blanchet
parents: 48311
diff changeset
   210
  | _ => ("", [])
48311
3c4e10606567 implemented MaSh QUERY operation
blanchet
parents: 48309
diff changeset
   211
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   212
fun mash_CLEAR ctxt =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   213
  let val path = mash_model_dir () in
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   214
    trace_msg ctxt (K "MaSh CLEAR");
50319
dddcaeb92e11 robust writing of MaSh state -- better drop learning data than cause other problems in Sledgehammer
blanchet
parents: 50311
diff changeset
   215
    try (File.fold_dir (fn file => fn _ =>
dddcaeb92e11 robust writing of MaSh state -- better drop learning data than cause other problems in Sledgehammer
blanchet
parents: 50311
diff changeset
   216
                           try File.rm (Path.append path (Path.basic file)))
dddcaeb92e11 robust writing of MaSh state -- better drop learning data than cause other problems in Sledgehammer
blanchet
parents: 50311
diff changeset
   217
                       path) NONE;
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   218
    ()
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   219
  end
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   220
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   221
fun mash_ADD _ _ [] = ()
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   222
  | mash_ADD ctxt overlord adds =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   223
    (trace_msg ctxt (fn () => "MaSh ADD " ^
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   224
         elide_string 1000 (space_implode " " (map #1 adds)));
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   225
     run_mash_tool ctxt overlord true 0 (adds, str_of_add) (K ()))
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   226
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   227
fun mash_REPROVE _ _ [] = ()
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   228
  | mash_REPROVE ctxt overlord reps =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   229
    (trace_msg ctxt (fn () => "MaSh REPROVE " ^
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   230
         elide_string 1000 (space_implode " " (map #1 reps)));
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   231
     run_mash_tool ctxt overlord true 0 (reps, str_of_reprove) (K ()))
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   232
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   233
fun mash_QUERY ctxt overlord max_suggs (query as (_, feats)) =
50356
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
   234
  (trace_msg ctxt (fn () => "MaSh QUERY " ^ encode_features feats);
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   235
   run_mash_tool ctxt overlord false max_suggs
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   236
       ([query], str_of_query)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   237
       (fn suggs =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   238
           case suggs () of
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   239
             [] => []
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   240
           | suggs => snd (extract_query (List.last suggs)))
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   241
   handle List.Empty => [])
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   242
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   243
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   244
(*** Middle-level communication with MaSh ***)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   245
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   246
datatype proof_kind = Isar_Proof | ATP_Proof | Isar_Proof_wegen_ATP_Flop
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   247
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   248
fun str_of_proof_kind Isar_Proof = "i"
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   249
  | str_of_proof_kind ATP_Proof = "a"
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   250
  | str_of_proof_kind Isar_Proof_wegen_ATP_Flop = "x"
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   251
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   252
fun proof_kind_of_str "i" = Isar_Proof
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   253
  | proof_kind_of_str "a" = ATP_Proof
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   254
  | proof_kind_of_str "x" = Isar_Proof_wegen_ATP_Flop
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   255
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   256
(* FIXME: Here a "Graph.update_node" function would be useful *)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   257
fun update_fact_graph_node (name, kind) =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   258
  Graph.default_node (name, Isar_Proof)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   259
  #> kind <> Isar_Proof ? Graph.map_node name (K kind)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   260
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   261
fun try_graph ctxt when def f =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   262
  f ()
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   263
  handle Graph.CYCLES (cycle :: _) =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   264
         (trace_msg ctxt (fn () =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   265
              "Cycle involving " ^ commas cycle ^ " when " ^ when); def)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   266
       | Graph.DUP name =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   267
         (trace_msg ctxt (fn () =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   268
              "Duplicate fact " ^ quote name ^ " when " ^ when); def)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   269
       | Graph.UNDEF name =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   270
         (trace_msg ctxt (fn () =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   271
              "Unknown fact " ^ quote name ^ " when " ^ when); def)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   272
       | exn =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   273
         if Exn.is_interrupt exn then
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   274
           reraise exn
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   275
         else
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   276
           (trace_msg ctxt (fn () =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   277
                "Internal error when " ^ when ^ ":\n" ^
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   278
                ML_Compiler.exn_message exn); def)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   279
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   280
fun graph_info G =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   281
  string_of_int (length (Graph.keys G)) ^ " node(s), " ^
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   282
  string_of_int (fold (Integer.add o length o snd) (Graph.dest G) 0) ^
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   283
  " edge(s), " ^
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   284
  string_of_int (length (Graph.minimals G)) ^ " minimal, " ^
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   285
  string_of_int (length (Graph.maximals G)) ^ " maximal"
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   286
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   287
type mash_state = {fact_G : unit Graph.T, dirty : string list option}
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   288
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   289
val empty_state = {fact_G = Graph.empty, dirty = SOME []}
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   290
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   291
local
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   292
50391
d2261e314146 tweaked MaSh features, based on comments by Josef Urban
blanchet
parents: 50390
diff changeset
   293
val version = "*** MaSh version 20121205c ***"
50357
187ae76a1757 more robustness in the face of MaSh format changes -- don't overwrite new versions with old versions
blanchet
parents: 50356
diff changeset
   294
187ae76a1757 more robustness in the face of MaSh format changes -- don't overwrite new versions with old versions
blanchet
parents: 50356
diff changeset
   295
exception Too_New of unit
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   296
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   297
fun extract_node line =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   298
  case space_explode ":" line of
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   299
    [head, parents] =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   300
    (case space_explode " " head of
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   301
       [kind, name] =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   302
       SOME (unescape_meta name, unescape_metas parents,
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   303
             try proof_kind_of_str kind |> the_default Isar_Proof)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   304
     | _ => NONE)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   305
  | _ => NONE
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   306
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   307
fun load _ (state as (true, _)) = state
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   308
  | load ctxt _ =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   309
    let val path = mash_state_file () in
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   310
      (true,
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   311
       case try File.read_lines path of
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   312
         SOME (version' :: node_lines) =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   313
         let
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   314
           fun add_edge_to name parent =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   315
             Graph.default_node (parent, Isar_Proof)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   316
             #> Graph.add_edge (parent, name)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   317
           fun add_node line =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   318
             case extract_node line of
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   319
               NONE => I (* shouldn't happen *)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   320
             | SOME (name, parents, kind) =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   321
               update_fact_graph_node (name, kind)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   322
               #> fold (add_edge_to name) parents
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   323
           val fact_G =
50357
187ae76a1757 more robustness in the face of MaSh format changes -- don't overwrite new versions with old versions
blanchet
parents: 50356
diff changeset
   324
             case string_ord (version', version) of
187ae76a1757 more robustness in the face of MaSh format changes -- don't overwrite new versions with old versions
blanchet
parents: 50356
diff changeset
   325
               EQUAL =>
187ae76a1757 more robustness in the face of MaSh format changes -- don't overwrite new versions with old versions
blanchet
parents: 50356
diff changeset
   326
               try_graph ctxt "loading state" Graph.empty (fn () =>
187ae76a1757 more robustness in the face of MaSh format changes -- don't overwrite new versions with old versions
blanchet
parents: 50356
diff changeset
   327
                   fold add_node node_lines Graph.empty)
187ae76a1757 more robustness in the face of MaSh format changes -- don't overwrite new versions with old versions
blanchet
parents: 50356
diff changeset
   328
             | LESS => Graph.empty (* can't parse old file *)
187ae76a1757 more robustness in the face of MaSh format changes -- don't overwrite new versions with old versions
blanchet
parents: 50356
diff changeset
   329
             | GREATER => raise Too_New ()
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   330
         in
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   331
           trace_msg ctxt (fn () =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   332
               "Loaded fact graph (" ^ graph_info fact_G ^ ")");
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   333
           {fact_G = fact_G, dirty = SOME []}
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   334
         end
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   335
       | _ => empty_state)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   336
    end
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   337
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   338
fun save _ (state as {dirty = SOME [], ...}) = state
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   339
  | save ctxt {fact_G, dirty} =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   340
    let
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   341
      fun str_of_entry (name, parents, kind) =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   342
        str_of_proof_kind kind ^ " " ^ escape_meta name ^ ": " ^
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   343
        escape_metas parents ^ "\n"
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   344
      fun append_entry (name, (kind, (parents, _))) =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   345
        cons (name, Graph.Keys.dest parents, kind)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   346
      val (banner, entries) =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   347
        case dirty of
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   348
          SOME names =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   349
          (NONE, fold (append_entry o Graph.get_entry fact_G) names [])
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   350
        | NONE => (SOME (version ^ "\n"), Graph.fold append_entry fact_G [])
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   351
    in
50335
b17b05c8d4a4 proper quoting of paths in MaSh shell script, take 2 (cf. b00eeb8e352e)
blanchet
parents: 50319
diff changeset
   352
      write_file banner (entries, str_of_entry) (mash_state_file ());
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   353
      trace_msg ctxt (fn () =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   354
          "Saved fact graph (" ^ graph_info fact_G ^
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   355
          (case dirty of
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   356
             SOME dirty =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   357
             "; " ^ string_of_int (length dirty) ^ " dirty fact(s)"
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   358
           | _ => "") ^  ")");
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   359
      {fact_G = fact_G, dirty = SOME []}
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   360
    end
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   361
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   362
val global_state =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   363
  Synchronized.var "Sledgehammer_MaSh.global_state" (false, empty_state)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   364
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   365
in
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   366
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   367
fun mash_map ctxt f =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   368
  Synchronized.change global_state (load ctxt ##> (f #> save ctxt))
50357
187ae76a1757 more robustness in the face of MaSh format changes -- don't overwrite new versions with old versions
blanchet
parents: 50356
diff changeset
   369
  handle Too_New () => ()
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   370
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   371
fun mash_peek ctxt f =
50357
187ae76a1757 more robustness in the face of MaSh format changes -- don't overwrite new versions with old versions
blanchet
parents: 50356
diff changeset
   372
  Synchronized.change_result global_state
187ae76a1757 more robustness in the face of MaSh format changes -- don't overwrite new versions with old versions
blanchet
parents: 50356
diff changeset
   373
      (perhaps (try (load ctxt)) #> `snd #>> f)
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   374
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   375
fun mash_get ctxt =
50357
187ae76a1757 more robustness in the face of MaSh format changes -- don't overwrite new versions with old versions
blanchet
parents: 50356
diff changeset
   376
  Synchronized.change_result global_state (perhaps (try (load ctxt)) #> `snd)
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   377
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   378
fun mash_unlearn ctxt =
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   379
  Synchronized.change global_state (fn _ =>
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   380
      (mash_CLEAR ctxt; (* also removes the state file *)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   381
       (true, empty_state)))
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   382
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   383
end
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   384
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   385
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   386
(*** Isabelle helpers ***)
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
   387
48378
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   388
fun parent_of_local_thm th =
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   389
  let
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   390
    val thy = th |> Thm.theory_of_thm
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   391
    val facts = thy |> Global_Theory.facts_of
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   392
    val space = facts |> Facts.space_of
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   393
    fun id_of s = #id (Name_Space.the_entry space s)
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   394
    fun max_id (s', _) (s, id) =
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   395
      let val id' = id_of s' in if id > id' then (s, id) else (s', id') end
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   396
  in ("", ~1) |> Facts.fold_static max_id facts |> fst end
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   397
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   398
val local_prefix = "local" ^ Long_Name.separator
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   399
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   400
fun nickname_of th =
48394
82fc8c956cdc fixed various issues with MaSh's file handling + tune output + generate local facts again + handle nameless facts gracefully
blanchet
parents: 48392
diff changeset
   401
  if Thm.has_name_hint th then
82fc8c956cdc fixed various issues with MaSh's file handling + tune output + generate local facts again + handle nameless facts gracefully
blanchet
parents: 48392
diff changeset
   402
    let val hint = Thm.get_name_hint th in
82fc8c956cdc fixed various issues with MaSh's file handling + tune output + generate local facts again + handle nameless facts gracefully
blanchet
parents: 48392
diff changeset
   403
      (* FIXME: There must be a better way to detect local facts. *)
82fc8c956cdc fixed various issues with MaSh's file handling + tune output + generate local facts again + handle nameless facts gracefully
blanchet
parents: 48392
diff changeset
   404
      case try (unprefix local_prefix) hint of
82fc8c956cdc fixed various issues with MaSh's file handling + tune output + generate local facts again + handle nameless facts gracefully
blanchet
parents: 48392
diff changeset
   405
        SOME suf =>
82fc8c956cdc fixed various issues with MaSh's file handling + tune output + generate local facts again + handle nameless facts gracefully
blanchet
parents: 48392
diff changeset
   406
        parent_of_local_thm th ^ Long_Name.separator ^ Long_Name.separator ^ suf
82fc8c956cdc fixed various issues with MaSh's file handling + tune output + generate local facts again + handle nameless facts gracefully
blanchet
parents: 48392
diff changeset
   407
      | NONE => hint
82fc8c956cdc fixed various issues with MaSh's file handling + tune output + generate local facts again + handle nameless facts gracefully
blanchet
parents: 48392
diff changeset
   408
    end
82fc8c956cdc fixed various issues with MaSh's file handling + tune output + generate local facts again + handle nameless facts gracefully
blanchet
parents: 48392
diff changeset
   409
  else
50047
45684acf0b94 thread context correctly when printing backquoted facts
blanchet
parents: 49997
diff changeset
   410
    backquote_thm (Proof_Context.init_global (Thm.theory_of_thm th)) th
48378
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   411
48330
192444b53e86 speed up MaSh queries
blanchet
parents: 48329
diff changeset
   412
fun suggested_facts suggs facts =
192444b53e86 speed up MaSh queries
blanchet
parents: 48329
diff changeset
   413
  let
48378
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   414
    fun add_fact (fact as (_, th)) = Symtab.default (nickname_of th, fact)
48330
192444b53e86 speed up MaSh queries
blanchet
parents: 48329
diff changeset
   415
    val tab = Symtab.empty |> fold add_fact facts
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   416
    fun find_sugg (name, weight) =
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   417
      Symtab.lookup tab name |> Option.map (rpair weight)
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   418
  in map_filter find_sugg suggs end
48311
3c4e10606567 implemented MaSh QUERY operation
blanchet
parents: 48309
diff changeset
   419
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   420
fun scaled_avg [] = 0
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   421
  | scaled_avg xs =
48407
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   422
    Real.ceil (100000000.0 * fold (curry (op +)) xs 0.0) div length xs
48328
ca0b7d19dd62 attempt at meshing according to more meaningful factors
blanchet
parents: 48327
diff changeset
   423
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   424
fun avg [] = 0.0
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   425
  | avg xs = fold (curry (op +)) xs 0.0 / Real.fromInt (length xs)
48313
0faafdffa662 mesh facts by taking into consideration whether a fact is known to MeSh
blanchet
parents: 48312
diff changeset
   426
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   427
fun normalize_scores _ [] = []
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   428
  | normalize_scores max_facts xs =
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   429
    let val avg = avg (map snd (take max_facts xs)) in
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   430
      map (apsnd (curry Real.* (1.0 / avg))) xs
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   431
    end
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   432
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   433
fun mesh_facts max_facts [(_, (sels, unks))] =
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   434
    map fst (take max_facts sels) @ take (max_facts - length sels) unks
48314
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
   435
  | mesh_facts max_facts mess =
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
   436
    let
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   437
      val mess =
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   438
        mess |> map (apsnd (apfst (normalize_scores max_facts #> `length)))
48314
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
   439
      val fact_eq = Thm.eq_thm o pairself snd
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   440
      fun score_in fact (global_weight, ((sel_len, sels), unks)) =
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   441
        let
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   442
          fun score_at j =
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   443
            case try (nth sels) j of
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   444
              SOME (_, score) => SOME (global_weight * score)
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   445
            | NONE => NONE
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   446
        in
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   447
          case find_index (curry fact_eq fact o fst) sels of
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   448
            ~1 => (case find_index (curry fact_eq fact) unks of
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   449
                     ~1 => score_at sel_len
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   450
                   | _ => NONE)
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   451
          | rank => score_at rank
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   452
        end
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   453
      fun weight_of fact = mess |> map_filter (score_in fact) |> scaled_avg
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   454
      val facts =
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   455
        fold (union fact_eq o map fst o take max_facts o snd o fst o snd) mess
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   456
             []
48314
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
   457
    in
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   458
      facts |> map (`weight_of) |> sort (int_ord o swap o pairself fst)
48328
ca0b7d19dd62 attempt at meshing according to more meaningful factors
blanchet
parents: 48327
diff changeset
   459
            |> map snd |> take max_facts
48314
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
   460
    end
48312
b40722a81ac9 implemented meshing of Iter and MaSh results
blanchet
parents: 48311
diff changeset
   461
50394
835a18063ed5 more MaSh feature tweaking
blanchet
parents: 50393
diff changeset
   462
fun thy_feature_of s = ("y" ^ s, 0.5 (* FUDGE *))
50393
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   463
fun const_feature_of s = ("c" ^ s, 1.0 (* FUDGE *))
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   464
fun free_feature_of s = ("f" ^ s, 2.0 (* FUDGE *))
50396
5c9a2f5ab323 more feature tweaks
blanchet
parents: 50395
diff changeset
   465
fun type_feature_of s = ("t" ^ s, 0.5 (* FUDGE *))
5c9a2f5ab323 more feature tweaks
blanchet
parents: 50395
diff changeset
   466
fun class_feature_of s = ("s" ^ s, 0.25 (* FUDGE *))
50394
835a18063ed5 more MaSh feature tweaking
blanchet
parents: 50393
diff changeset
   467
fun status_feature_of status = (string_of_status status, 0.5 (* FUDGE *))
50390
b7b1f2a170b8 increase weight of local facts again (MaSh)
blanchet
parents: 50389
diff changeset
   468
val local_feature = ("local", 2.0 (* FUDGE *))
50391
d2261e314146 tweaked MaSh features, based on comments by Josef Urban
blanchet
parents: 50390
diff changeset
   469
val lams_feature = ("lams", 0.5 (* FUDGE *))
d2261e314146 tweaked MaSh features, based on comments by Josef Urban
blanchet
parents: 50390
diff changeset
   470
val skos_feature = ("skos", 0.5 (* FUDGE *))
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   471
48324
3ee5b5589402 speed up tautology/metaness check
blanchet
parents: 48323
diff changeset
   472
fun theory_ord p =
3ee5b5589402 speed up tautology/metaness check
blanchet
parents: 48323
diff changeset
   473
  if Theory.eq_thy p then
3ee5b5589402 speed up tautology/metaness check
blanchet
parents: 48323
diff changeset
   474
    EQUAL
3ee5b5589402 speed up tautology/metaness check
blanchet
parents: 48323
diff changeset
   475
  else if Theory.subthy p then
3ee5b5589402 speed up tautology/metaness check
blanchet
parents: 48323
diff changeset
   476
    LESS
3ee5b5589402 speed up tautology/metaness check
blanchet
parents: 48323
diff changeset
   477
  else if Theory.subthy (swap p) then
3ee5b5589402 speed up tautology/metaness check
blanchet
parents: 48323
diff changeset
   478
    GREATER
3ee5b5589402 speed up tautology/metaness check
blanchet
parents: 48323
diff changeset
   479
  else case int_ord (pairself (length o Theory.ancestors_of) p) of
3ee5b5589402 speed up tautology/metaness check
blanchet
parents: 48323
diff changeset
   480
    EQUAL => string_ord (pairself Context.theory_name p)
3ee5b5589402 speed up tautology/metaness check
blanchet
parents: 48323
diff changeset
   481
  | order => order
3ee5b5589402 speed up tautology/metaness check
blanchet
parents: 48323
diff changeset
   482
50382
blanchet
parents: 50359
diff changeset
   483
fun thm_ord p =
blanchet
parents: 50359
diff changeset
   484
  case theory_ord (pairself theory_of_thm p) of
50359
da395f0e7dea tweaked order of theorems to avoid forward dependencies (MaSh)
blanchet
parents: 50357
diff changeset
   485
    EQUAL =>
da395f0e7dea tweaked order of theorems to avoid forward dependencies (MaSh)
blanchet
parents: 50357
diff changeset
   486
    (* Hack to put "xxx_def" before "xxxI" and "xxxE" *)
50382
blanchet
parents: 50359
diff changeset
   487
    string_ord (pairself nickname_of (swap p))
50359
da395f0e7dea tweaked order of theorems to avoid forward dependencies (MaSh)
blanchet
parents: 50357
diff changeset
   488
  | ord => ord
48324
3ee5b5589402 speed up tautology/metaness check
blanchet
parents: 48323
diff changeset
   489
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   490
val freezeT = Type.legacy_freeze_type
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   491
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   492
fun freeze (t $ u) = freeze t $ freeze u
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   493
  | freeze (Abs (s, T, t)) = Abs (s, freezeT T, freeze t)
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   494
  | freeze (Var ((s, _), T)) = Free (s, freezeT T)
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   495
  | freeze (Const (s, T)) = Const (s, freezeT T)
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   496
  | freeze (Free (s, T)) = Free (s, freezeT T)
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   497
  | freeze t = t
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   498
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   499
fun goal_of_thm thy = prop_of #> freeze #> cterm_of thy #> Goal.init
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   500
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   501
fun run_prover_for_mash ctxt params prover facts goal =
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   502
  let
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   503
    val problem =
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   504
      {state = Proof.init ctxt, goal = goal, subgoal = 1, subgoal_count = 1,
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   505
       facts = facts |> map (apfst (apfst (fn name => name ())))
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   506
                     |> map Untranslated_Fact}
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   507
  in
48399
4bacc8983b3d learn from SMT proofs when they can be minimized by Metis
blanchet
parents: 48398
diff changeset
   508
    get_minimizing_prover ctxt MaSh (K (K ())) prover params (K (K (K "")))
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   509
                          problem
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   510
  end
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   511
48326
ef800e91d072 removed debugging output
blanchet
parents: 48325
diff changeset
   512
val bad_types = [@{type_name prop}, @{type_name bool}, @{type_name fun}]
ef800e91d072 removed debugging output
blanchet
parents: 48325
diff changeset
   513
48398
b86450f5b5cb clean up interesting constants a bit
blanchet
parents: 48396
diff changeset
   514
val logical_consts =
b86450f5b5cb clean up interesting constants a bit
blanchet
parents: 48396
diff changeset
   515
  [@{const_name prop}, @{const_name Pure.conjunction}] @ atp_logical_consts
b86450f5b5cb clean up interesting constants a bit
blanchet
parents: 48396
diff changeset
   516
50393
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   517
fun interesting_terms_types_and_classes ctxt thy_name prover term_max_depth
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
   518
                                        type_max_depth ts =
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   519
  let
50392
190053ee24ed expand type classes into their ancestors for MaSh
blanchet
parents: 50391
diff changeset
   520
    val thy = Proof_Context.theory_of ctxt
50393
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   521
    val fixes = map snd (Variable.dest_fixes ctxt)
50392
190053ee24ed expand type classes into their ancestors for MaSh
blanchet
parents: 50391
diff changeset
   522
    val classes = Sign.classes_of thy
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
   523
    fun is_bad_const (x as (s, _)) args =
48398
b86450f5b5cb clean up interesting constants a bit
blanchet
parents: 48396
diff changeset
   524
      member (op =) logical_consts s orelse
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
   525
      fst (is_built_in_const_for_prover ctxt prover x args)
48304
50e64af9c829 more work on MaSh
blanchet
parents: 48303
diff changeset
   526
    fun add_classes @{sort type} = I
50392
190053ee24ed expand type classes into their ancestors for MaSh
blanchet
parents: 50391
diff changeset
   527
      | add_classes S =
190053ee24ed expand type classes into their ancestors for MaSh
blanchet
parents: 50391
diff changeset
   528
        fold (`(Sorts.super_classes classes)
190053ee24ed expand type classes into their ancestors for MaSh
blanchet
parents: 50391
diff changeset
   529
              #> swap #> op ::
190053ee24ed expand type classes into their ancestors for MaSh
blanchet
parents: 50391
diff changeset
   530
              #> subtract (op =) @{sort type}
190053ee24ed expand type classes into their ancestors for MaSh
blanchet
parents: 50391
diff changeset
   531
              #> map class_feature_of
190053ee24ed expand type classes into their ancestors for MaSh
blanchet
parents: 50391
diff changeset
   532
              #> union (op = o pairself fst)) S
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   533
    fun do_add_type (Type (s, Ts)) =
50356
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
   534
        (not (member (op =) bad_types s)
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
   535
         ? insert (op = o pairself fst) (type_feature_of s))
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   536
        #> fold do_add_type Ts
48304
50e64af9c829 more work on MaSh
blanchet
parents: 48303
diff changeset
   537
      | do_add_type (TFree (_, S)) = add_classes S
50e64af9c829 more work on MaSh
blanchet
parents: 48303
diff changeset
   538
      | do_add_type (TVar (_, S)) = add_classes S
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   539
    fun add_type T = type_max_depth >= 0 ? do_add_type T
50339
d8dae91f3107 MaSh improvements: deeper patterns + more respect for chained facts
blanchet
parents: 50338
diff changeset
   540
    fun patternify_term _ ~1 _ = []
d8dae91f3107 MaSh improvements: deeper patterns + more respect for chained facts
blanchet
parents: 50338
diff changeset
   541
      | patternify_term args _ (Const (x as (s, _))) =
50356
41f8f1f2e15d added feature weights in MaSh
blanchet
parents: 50355
diff changeset
   542
        if is_bad_const x args then [] else [s]
50393
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   543
      | patternify_term _ depth (Free (s, _)) =
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   544
        if depth = term_max_depth andalso member (op =) fixes s then
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   545
          [thy_name ^ Long_Name.separator ^ s]
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   546
        else
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   547
          []
50339
d8dae91f3107 MaSh improvements: deeper patterns + more respect for chained facts
blanchet
parents: 50338
diff changeset
   548
      | patternify_term _ 0 _ = []
d8dae91f3107 MaSh improvements: deeper patterns + more respect for chained facts
blanchet
parents: 50338
diff changeset
   549
      | patternify_term args depth (t $ u) =
d8dae91f3107 MaSh improvements: deeper patterns + more respect for chained facts
blanchet
parents: 50338
diff changeset
   550
        let
d8dae91f3107 MaSh improvements: deeper patterns + more respect for chained facts
blanchet
parents: 50338
diff changeset
   551
          val ps = patternify_term (u :: args) depth t
d8dae91f3107 MaSh improvements: deeper patterns + more respect for chained facts
blanchet
parents: 50338
diff changeset
   552
          val qs = "" :: patternify_term [] (depth - 1) u
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   553
        in map_product (fn p => fn "" => p | q => p ^ "(" ^ q ^ ")") ps qs end
50339
d8dae91f3107 MaSh improvements: deeper patterns + more respect for chained facts
blanchet
parents: 50338
diff changeset
   554
      | patternify_term _ _ _ = []
50393
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   555
    fun add_term_pattern feature_of =
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   556
      union (op = o pairself fst) o map feature_of oo patternify_term []
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   557
    fun add_term_patterns _ ~1 _ = I
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   558
      | add_term_patterns feature_of depth t =
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   559
        add_term_pattern feature_of depth t
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   560
        #> add_term_patterns feature_of (depth - 1) t
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   561
    fun add_term feature_of = add_term_patterns feature_of term_max_depth
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   562
    fun add_patterns t =
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   563
      let val (head, args) = strip_comb t in
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   564
        (case head of
50393
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   565
           Const (_, T) => add_term const_feature_of t #> add_type T
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   566
         | Free (_, T) => add_term free_feature_of t #> add_type T
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   567
         | Var (_, T) => add_type T
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   568
         | Abs (_, T, body) => add_type T #> add_patterns body
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   569
         | _ => I)
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   570
        #> fold add_patterns args
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   571
      end
48326
ef800e91d072 removed debugging output
blanchet
parents: 48325
diff changeset
   572
  in [] |> fold add_patterns ts end
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   573
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   574
fun is_exists (s, _) = (s = @{const_name Ex} orelse s = @{const_name Ex1})
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   575
50396
5c9a2f5ab323 more feature tweaks
blanchet
parents: 50395
diff changeset
   576
val term_max_depth = 1
5c9a2f5ab323 more feature tweaks
blanchet
parents: 50395
diff changeset
   577
val type_max_depth = 1
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   578
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   579
(* TODO: Generate type classes for types? *)
48385
2779dea0b1e0 added locality as a MaSh feature
blanchet
parents: 48384
diff changeset
   580
fun features_of ctxt prover thy (scope, status) ts =
50393
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   581
  let val thy_name = Context.theory_name thy in
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   582
    thy_feature_of thy_name ::
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   583
    interesting_terms_types_and_classes ctxt thy_name prover term_max_depth
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   584
        type_max_depth ts
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   585
    |> status <> General ? cons (status_feature_of status)
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   586
    |> scope <> Global ? cons local_feature
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   587
    |> exists (not o is_lambda_free) ts ? cons lams_feature
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   588
    |> exists (exists_Const is_exists) ts ? cons skos_feature
d8aa26c78327 record free variables as a MaSh feature
blanchet
parents: 50392
diff changeset
   589
  end
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   590
48665
14b0732c72f7 crank up max number of dependencies
blanchet
parents: 48656
diff changeset
   591
(* Too many dependencies is a sign that a crazy decision procedure is at work.
14b0732c72f7 crank up max number of dependencies
blanchet
parents: 48656
diff changeset
   592
   There isn't much to learn from such proofs. *)
50397
d84a5ab736bb reduce max number of dependencies for MaSh to get rid of junk
blanchet
parents: 50396
diff changeset
   593
val max_dependencies = 10
48665
14b0732c72f7 crank up max number of dependencies
blanchet
parents: 48656
diff changeset
   594
val atp_dependency_default_max_facts = 50
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   595
48438
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   596
(* "type_definition_xxx" facts are characterized by their use of "CollectI". *)
48441
2d2f009ca8eb remove MaSh junk associated with size functions
blanchet
parents: 48440
diff changeset
   597
val typedef_deps = [@{thm CollectI} |> nickname_of]
48438
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   598
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   599
(* "Rep_xxx_inject", "Abs_xxx_inverse", etc., are derived using these facts. *)
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   600
val typedef_ths =
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   601
  @{thms type_definition.Abs_inverse type_definition.Rep_inverse
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   602
         type_definition.Rep type_definition.Rep_inject
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   603
         type_definition.Abs_inject type_definition.Rep_cases
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   604
         type_definition.Abs_cases type_definition.Rep_induct
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   605
         type_definition.Abs_induct type_definition.Rep_range
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   606
         type_definition.Abs_image}
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   607
  |> map nickname_of
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   608
48441
2d2f009ca8eb remove MaSh junk associated with size functions
blanchet
parents: 48440
diff changeset
   609
fun is_size_def [dep] th =
2d2f009ca8eb remove MaSh junk associated with size functions
blanchet
parents: 48440
diff changeset
   610
    (case first_field ".recs" dep of
2d2f009ca8eb remove MaSh junk associated with size functions
blanchet
parents: 48440
diff changeset
   611
       SOME (pref, _) =>
2d2f009ca8eb remove MaSh junk associated with size functions
blanchet
parents: 48440
diff changeset
   612
       (case first_field ".size" (nickname_of th) of
2d2f009ca8eb remove MaSh junk associated with size functions
blanchet
parents: 48440
diff changeset
   613
          SOME (pref', _) => pref = pref'
2d2f009ca8eb remove MaSh junk associated with size functions
blanchet
parents: 48440
diff changeset
   614
        | NONE => false)
2d2f009ca8eb remove MaSh junk associated with size functions
blanchet
parents: 48440
diff changeset
   615
     | NONE => false)
2d2f009ca8eb remove MaSh junk associated with size functions
blanchet
parents: 48440
diff changeset
   616
  | is_size_def _ _ = false
2d2f009ca8eb remove MaSh junk associated with size functions
blanchet
parents: 48440
diff changeset
   617
2d2f009ca8eb remove MaSh junk associated with size functions
blanchet
parents: 48440
diff changeset
   618
fun trim_dependencies th deps =
48668
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   619
  if length deps > max_dependencies then
48438
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   620
    NONE
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48436
diff changeset
   621
  else
48668
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   622
    SOME (if deps = typedef_deps orelse
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   623
             exists (member (op =) typedef_ths) deps orelse
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   624
             is_size_def deps th then
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   625
            []
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   626
          else
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   627
            deps)
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   628
48441
2d2f009ca8eb remove MaSh junk associated with size functions
blanchet
parents: 48440
diff changeset
   629
fun isar_dependencies_of all_names th =
2d2f009ca8eb remove MaSh junk associated with size functions
blanchet
parents: 48440
diff changeset
   630
  th |> thms_in_proof (SOME all_names) |> trim_dependencies th
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   631
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   632
fun atp_dependencies_of ctxt (params as {verbose, max_facts, ...}) prover
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   633
                        auto_level facts all_names th =
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   634
  case isar_dependencies_of all_names th of
48665
14b0732c72f7 crank up max number of dependencies
blanchet
parents: 48656
diff changeset
   635
    SOME [] => (false, SOME [])
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   636
  | isar_deps =>
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   637
    let
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   638
      val thy = Proof_Context.theory_of ctxt
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   639
      val goal = goal_of_thm thy th
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   640
      val (_, hyp_ts, concl_t) = ATP_Util.strip_subgoal ctxt goal 1
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   641
      val facts = facts |> filter (fn (_, th') => thm_ord (th', th) = LESS)
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   642
      fun fix_name ((_, stature), th) = ((fn () => nickname_of th, stature), th)
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   643
      fun is_dep dep (_, th) = nickname_of th = dep
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   644
      fun add_isar_dep facts dep accum =
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   645
        if exists (is_dep dep) accum then
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   646
          accum
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   647
        else case find_first (is_dep dep) facts of
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   648
          SOME ((name, status), th) => accum @ [((name, status), th)]
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   649
        | NONE => accum (* shouldn't happen *)
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   650
      val facts =
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   651
        facts |> mepo_suggested_facts ctxt params prover
48665
14b0732c72f7 crank up max number of dependencies
blanchet
parents: 48656
diff changeset
   652
                     (max_facts |> the_default atp_dependency_default_max_facts)
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   653
                     NONE hyp_ts concl_t
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   654
              |> fold (add_isar_dep facts) (these isar_deps)
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   655
              |> map fix_name
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   656
    in
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   657
      if verbose andalso auto_level = 0 then
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   658
        let val num_facts = length facts in
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   659
          "MaSh: " ^ quote prover ^ " on " ^ quote (nickname_of th) ^
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   660
          " with " ^ string_of_int num_facts ^ " fact" ^ plural_s num_facts ^
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   661
          "."
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   662
          |> Output.urgent_message
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   663
        end
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   664
      else
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   665
        ();
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   666
      case run_prover_for_mash ctxt params prover facts goal of
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   667
        {outcome = NONE, used_facts, ...} =>
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   668
        (if verbose andalso auto_level = 0 then
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   669
           let val num_facts = length used_facts in
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   670
             "Found proof with " ^ string_of_int num_facts ^ " fact" ^
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   671
             plural_s num_facts ^ "."
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   672
             |> Output.urgent_message
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   673
           end
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   674
         else
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   675
           ();
48665
14b0732c72f7 crank up max number of dependencies
blanchet
parents: 48656
diff changeset
   676
         case used_facts |> map fst |> trim_dependencies th of
14b0732c72f7 crank up max number of dependencies
blanchet
parents: 48656
diff changeset
   677
           NONE => (false, isar_deps)
14b0732c72f7 crank up max number of dependencies
blanchet
parents: 48656
diff changeset
   678
         | atp_deps => (true, atp_deps))
14b0732c72f7 crank up max number of dependencies
blanchet
parents: 48656
diff changeset
   679
      | _ => (false, isar_deps)
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   680
    end
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   681
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   682
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   683
(*** High-level communication with MaSh ***)
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48249
diff changeset
   684
48407
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   685
fun num_keys keys = Graph.Keys.fold (K (Integer.add 1)) keys 0
48400
f08425165cca minimal maxes + tuning
blanchet
parents: 48399
diff changeset
   686
48407
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   687
fun maximal_in_graph fact_G facts =
48316
252f45c04042 drastic overhaul of MaSh data structures + fixed a few performance issues
blanchet
parents: 48315
diff changeset
   688
  let
48378
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   689
    val facts = [] |> fold (cons o nickname_of o snd) facts
48407
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   690
    val tab = Symtab.empty |> fold (fn name => Symtab.default (name, ())) facts
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   691
    fun insert_new seen name =
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   692
      not (Symtab.defined seen name) ? insert (op =) name
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   693
    fun find_maxes _ (maxs, []) = map snd maxs
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   694
      | find_maxes seen (maxs, new :: news) =
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   695
        find_maxes
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   696
            (seen |> num_keys (Graph.imm_succs fact_G new) > 1
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   697
                     ? Symtab.default (new, ()))
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   698
            (if Symtab.defined tab new then
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   699
               let
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   700
                 val newp = Graph.all_preds fact_G [new]
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   701
                 fun is_ancestor x yp = member (op =) yp x
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   702
                 val maxs =
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   703
                   maxs |> filter (fn (_, max) => not (is_ancestor max newp))
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   704
               in
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   705
                 if exists (is_ancestor new o fst) maxs then
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   706
                   (maxs, news)
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   707
                 else
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   708
                   ((newp, new)
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   709
                    :: filter_out (fn (_, max) => is_ancestor max newp) maxs,
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   710
                    news)
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   711
               end
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   712
             else
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   713
               (maxs, Graph.Keys.fold (insert_new seen)
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   714
                                      (Graph.imm_preds fact_G new) news))
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   715
  in find_maxes Symtab.empty ([], Graph.maximals fact_G) end
48316
252f45c04042 drastic overhaul of MaSh data structures + fixed a few performance issues
blanchet
parents: 48315
diff changeset
   716
48400
f08425165cca minimal maxes + tuning
blanchet
parents: 48399
diff changeset
   717
fun is_fact_in_graph fact_G (_, th) =
f08425165cca minimal maxes + tuning
blanchet
parents: 48399
diff changeset
   718
  can (Graph.get_node fact_G) (nickname_of th)
48320
891a24a48155 improved meshing of MaSh and Meng--Paulson if some MaSh suggestions are cut-off (the common case)
blanchet
parents: 48319
diff changeset
   719
50340
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
   720
(* factor that controls whether unknown global facts should be included *)
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
   721
val include_unk_global_factor = 15
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
   722
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   723
(* use MePo weights for now *)
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   724
val weight_raw_mash_facts = weight_mepo_facts
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   725
val weight_mash_facts = weight_raw_mash_facts
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   726
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   727
(* FUDGE *)
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   728
fun weight_of_proximity_fact rank =
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   729
  Math.pow (1.3, 15.5 - 0.05 * Real.fromInt rank) + 15.0
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   730
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   731
fun weight_proximity_facts facts =
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   732
  facts ~~ map weight_of_proximity_fact (0 upto length facts - 1)
50382
blanchet
parents: 50359
diff changeset
   733
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   734
fun mash_suggested_facts ctxt ({overlord, ...} : params) prover max_facts hyp_ts
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   735
                         concl_t facts =
48301
e5c5037a3104 started implementing MaSh client-side I/O
blanchet
parents: 48299
diff changeset
   736
  let
48302
6cf5e58f1185 more implementation work on MaSh
blanchet
parents: 48301
diff changeset
   737
    val thy = Proof_Context.theory_of ctxt
48434
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   738
    val (fact_G, suggs) =
48699
a89b83204c24 optimized saving
blanchet
parents: 48669
diff changeset
   739
      mash_peek ctxt (fn {fact_G, ...} =>
48434
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   740
          if Graph.is_empty fact_G then
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   741
            (fact_G, [])
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   742
          else
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   743
            let
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   744
              val parents = maximal_in_graph fact_G facts
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   745
              val feats =
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   746
                features_of ctxt prover thy (Local, General) (concl_t :: hyp_ts)
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   747
            in
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   748
              (fact_G, mash_QUERY ctxt overlord max_facts (parents, feats))
48434
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   749
            end)
50397
d84a5ab736bb reduce max number of dependencies for MaSh to get rid of junk
blanchet
parents: 50396
diff changeset
   750
    val chained = facts |> filter (fn ((_, (scope, _)), _) => scope = Chained)
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   751
    val raw_mash =
48408
5493e67982ee tune Mesh filter
blanchet
parents: 48407
diff changeset
   752
      facts |> suggested_facts suggs
50339
d8dae91f3107 MaSh improvements: deeper patterns + more respect for chained facts
blanchet
parents: 50338
diff changeset
   753
            (* The weights currently returned by "mash.py" are too spaced out to
48408
5493e67982ee tune Mesh filter
blanchet
parents: 48407
diff changeset
   754
               make any sense. *)
48435
f30eb5eb7927 include unknown local facts in MaSh
blanchet
parents: 48434
diff changeset
   755
            |> map fst
50395
c69a970143c0 prioritize chained facts
blanchet
parents: 50394
diff changeset
   756
    val proximity = facts |> sort (thm_ord o pairself snd o swap)
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   757
    val unknown = facts |> filter_out (is_fact_in_graph fact_G)
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   758
    val mess =
50395
c69a970143c0 prioritize chained facts
blanchet
parents: 50394
diff changeset
   759
      [(0.8000 (* FUDGE *), (map (rpair 1.0) chained, [])),
c69a970143c0 prioritize chained facts
blanchet
parents: 50394
diff changeset
   760
       (0.1333 (* FUDGE *), (weight_raw_mash_facts raw_mash, unknown)),
c69a970143c0 prioritize chained facts
blanchet
parents: 50394
diff changeset
   761
       (0.0667 (* FUDGE *), (weight_proximity_facts proximity, []))]
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
   762
  in mesh_facts max_facts mess end
48249
2bd242c56c90 dummy implementation
blanchet
parents: 48248
diff changeset
   763
48668
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   764
fun add_wrt_fact_graph ctxt (name, parents, feats, deps) (adds, graph) =
48316
252f45c04042 drastic overhaul of MaSh data structures + fixed a few performance issues
blanchet
parents: 48315
diff changeset
   765
  let
252f45c04042 drastic overhaul of MaSh data structures + fixed a few performance issues
blanchet
parents: 48315
diff changeset
   766
    fun maybe_add_from from (accum as (parents, graph)) =
48321
c552d7f1720b learn from minimized ATP proofs
blanchet
parents: 48320
diff changeset
   767
      try_graph ctxt "updating graph" accum (fn () =>
c552d7f1720b learn from minimized ATP proofs
blanchet
parents: 48320
diff changeset
   768
          (from :: parents, Graph.add_edge_acyclic (from, name) graph))
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   769
    val graph = graph |> Graph.default_node (name, Isar_Proof)
48316
252f45c04042 drastic overhaul of MaSh data structures + fixed a few performance issues
blanchet
parents: 48315
diff changeset
   770
    val (parents, graph) = ([], graph) |> fold maybe_add_from parents
48407
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   771
    val (deps, _) = ([], graph) |> fold maybe_add_from deps
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   772
  in ((name, parents, feats, deps) :: adds, graph) end
48306
e699f754d9f7 better zipping of MaSh facts
blanchet
parents: 48305
diff changeset
   773
48668
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   774
fun reprove_wrt_fact_graph ctxt (name, deps) (reps, graph) =
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   775
  let
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   776
    fun maybe_rep_from from (accum as (parents, graph)) =
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   777
      try_graph ctxt "updating graph" accum (fn () =>
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   778
          (from :: parents, Graph.add_edge_acyclic (from, name) graph))
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   779
    val graph = graph |> update_fact_graph_node (name, ATP_Proof)
48668
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   780
    val (deps, _) = ([], graph) |> fold maybe_rep_from deps
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   781
  in ((name, deps) :: reps, graph) end
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   782
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   783
fun flop_wrt_fact_graph name =
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   784
  update_fact_graph_node (name, Isar_Proof_wegen_ATP_Flop)
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   785
48384
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
   786
val learn_timeout_slack = 2.0
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
   787
48384
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
   788
fun launch_thread timeout task =
48383
df75b2d7e26a learn command in MaSh
blanchet
parents: 48382
diff changeset
   789
  let
48384
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
   790
    val hard_timeout = time_mult learn_timeout_slack timeout
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
   791
    val birth_time = Time.now ()
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
   792
    val death_time = Time.+ (birth_time, hard_timeout)
48442
3c9890c19e90 cap the number of facts returned by MaSh
blanchet
parents: 48441
diff changeset
   793
    val desc = ("Machine learner for Sledgehammer", "")
48384
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
   794
  in Async_Manager.launch MaShN birth_time death_time desc task end
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
   795
48400
f08425165cca minimal maxes + tuning
blanchet
parents: 48399
diff changeset
   796
fun freshish_name () =
f08425165cca minimal maxes + tuning
blanchet
parents: 48399
diff changeset
   797
  Date.fmt ".%Y_%m_%d_%H_%M_%S__" (Date.fromTimeLocal (Time.now ())) ^
f08425165cca minimal maxes + tuning
blanchet
parents: 48399
diff changeset
   798
  serial_string ()
f08425165cca minimal maxes + tuning
blanchet
parents: 48399
diff changeset
   799
48384
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
   800
fun mash_learn_proof ctxt ({overlord, timeout, ...} : params) prover t facts
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
   801
                     used_ths =
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
   802
  if is_smt_prover ctxt prover then
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
   803
    ()
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
   804
  else
48403
1f214c653c80 don't store fresh names in fact graph, since these cannot be the parents of any other facts
blanchet
parents: 48401
diff changeset
   805
    launch_thread timeout (fn () =>
1f214c653c80 don't store fresh names in fact graph, since these cannot be the parents of any other facts
blanchet
parents: 48401
diff changeset
   806
        let
1f214c653c80 don't store fresh names in fact graph, since these cannot be the parents of any other facts
blanchet
parents: 48401
diff changeset
   807
          val thy = Proof_Context.theory_of ctxt
1f214c653c80 don't store fresh names in fact graph, since these cannot be the parents of any other facts
blanchet
parents: 48401
diff changeset
   808
          val name = freshish_name ()
1f214c653c80 don't store fresh names in fact graph, since these cannot be the parents of any other facts
blanchet
parents: 48401
diff changeset
   809
          val feats = features_of ctxt prover thy (Local, General) [t]
1f214c653c80 don't store fresh names in fact graph, since these cannot be the parents of any other facts
blanchet
parents: 48401
diff changeset
   810
          val deps = used_ths |> map nickname_of
1f214c653c80 don't store fresh names in fact graph, since these cannot be the parents of any other facts
blanchet
parents: 48401
diff changeset
   811
        in
48699
a89b83204c24 optimized saving
blanchet
parents: 48669
diff changeset
   812
          mash_peek ctxt (fn {fact_G, ...} =>
48434
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   813
              let val parents = maximal_in_graph fact_G facts in
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   814
                mash_ADD ctxt overlord [(name, parents, feats, deps)]
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   815
              end);
aaaec69db3db ensure all calls to "mash" program are synchronous
blanchet
parents: 48433
diff changeset
   816
          (true, "")
48403
1f214c653c80 don't store fresh names in fact graph, since these cannot be the parents of any other facts
blanchet
parents: 48401
diff changeset
   817
        end)
48383
df75b2d7e26a learn command in MaSh
blanchet
parents: 48382
diff changeset
   818
50163
c62ce309dc26 more abstract Sendback operations, with explicit id/exec_id properties;
wenzelm
parents: 50047
diff changeset
   819
fun sendback sub = Sendback.markup (sledgehammerN ^ " " ^ sub)
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   820
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   821
val commit_timeout = seconds 30.0
48332
271a4a6af734 optimize parent computation in MaSh + remove temporary files
blanchet
parents: 48330
diff changeset
   822
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
   823
(* The timeout is understood in a very slack fashion. *)
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   824
fun mash_learn_facts ctxt (params as {debug, verbose, overlord, ...}) prover
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   825
                     auto_level atp learn_timeout facts =
48304
50e64af9c829 more work on MaSh
blanchet
parents: 48303
diff changeset
   826
  let
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
   827
    val timer = Timer.startRealTimer ()
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   828
    fun next_commit_time () =
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   829
      Time.+ (Timer.checkRealTimer timer, commit_timeout)
48699
a89b83204c24 optimized saving
blanchet
parents: 48669
diff changeset
   830
    val {fact_G, ...} = mash_get ctxt
48400
f08425165cca minimal maxes + tuning
blanchet
parents: 48399
diff changeset
   831
    val (old_facts, new_facts) =
48667
ac58317ef11f rule out same "technical" theories for MePo as for MaSh
blanchet
parents: 48665
diff changeset
   832
      facts |> List.partition (is_fact_in_graph fact_G)
48400
f08425165cca minimal maxes + tuning
blanchet
parents: 48399
diff changeset
   833
            ||> sort (thm_ord o pairself snd)
48308
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
   834
  in
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   835
    if null new_facts andalso (not atp orelse null old_facts) then
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   836
      if auto_level < 2 then
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   837
        "No new " ^ (if atp then "ATP" else "Isar") ^ " proofs to learn." ^
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   838
        (if auto_level = 0 andalso not atp then
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   839
           "\n\nHint: Try " ^ sendback learn_atpN ^ " to learn from ATP proofs."
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   840
         else
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   841
           "")
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   842
      else
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   843
        ""
48308
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
   844
    else
48304
50e64af9c829 more work on MaSh
blanchet
parents: 48303
diff changeset
   845
      let
48315
82d6e46c673f fixed order of accessibles + other tweaks to MaSh
blanchet
parents: 48314
diff changeset
   846
        val all_names =
49997
dbbf9578e712 made MaSh more robust in the face of duplicate "nicknames" (which can happen e.g. if you have a lemma called foo(1) and another called foo_1 in the same theory)
blanchet
parents: 48699
diff changeset
   847
          Symtab.empty
dbbf9578e712 made MaSh more robust in the face of duplicate "nicknames" (which can happen e.g. if you have a lemma called foo(1) and another called foo_1 in the same theory)
blanchet
parents: 48699
diff changeset
   848
          |> fold Symtab.update (facts |> map (rpair () o nickname_of o snd))
48439
67a6bcbd3587 removed MaSh junk arising from primrec definitions
blanchet
parents: 48438
diff changeset
   849
        fun deps_of status th =
67a6bcbd3587 removed MaSh junk arising from primrec definitions
blanchet
parents: 48438
diff changeset
   850
          if status = Non_Rec_Def orelse status = Rec_Def then
67a6bcbd3587 removed MaSh junk arising from primrec definitions
blanchet
parents: 48438
diff changeset
   851
            SOME []
67a6bcbd3587 removed MaSh junk arising from primrec definitions
blanchet
parents: 48438
diff changeset
   852
          else if atp then
67a6bcbd3587 removed MaSh junk arising from primrec definitions
blanchet
parents: 48438
diff changeset
   853
            atp_dependencies_of ctxt params prover auto_level facts all_names th
48665
14b0732c72f7 crank up max number of dependencies
blanchet
parents: 48656
diff changeset
   854
            |> (fn (false, _) => NONE | (true, deps) => deps)
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   855
          else
48439
67a6bcbd3587 removed MaSh junk arising from primrec definitions
blanchet
parents: 48438
diff changeset
   856
            isar_dependencies_of all_names th
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   857
        fun do_commit [] [] [] state = state
48699
a89b83204c24 optimized saving
blanchet
parents: 48669
diff changeset
   858
          | do_commit adds reps flops {fact_G, dirty} =
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   859
            let
48699
a89b83204c24 optimized saving
blanchet
parents: 48669
diff changeset
   860
              val was_empty = Graph.is_empty fact_G
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   861
              val (adds, fact_G) =
48668
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   862
                ([], fact_G) |> fold (add_wrt_fact_graph ctxt) adds
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   863
              val (reps, fact_G) =
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   864
                ([], fact_G) |> fold (reprove_wrt_fact_graph ctxt) reps
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   865
              val fact_G = fact_G |> fold flop_wrt_fact_graph flops
48699
a89b83204c24 optimized saving
blanchet
parents: 48669
diff changeset
   866
              val dirty =
a89b83204c24 optimized saving
blanchet
parents: 48669
diff changeset
   867
                case (was_empty, dirty, reps) of
a89b83204c24 optimized saving
blanchet
parents: 48669
diff changeset
   868
                  (false, SOME names, []) => SOME (map #1 adds @ names)
a89b83204c24 optimized saving
blanchet
parents: 48669
diff changeset
   869
                | _ => NONE
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   870
            in
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   871
              mash_ADD ctxt overlord (rev adds);
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   872
              mash_REPROVE ctxt overlord reps;
48699
a89b83204c24 optimized saving
blanchet
parents: 48669
diff changeset
   873
              {fact_G = fact_G, dirty = dirty}
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   874
            end
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   875
        fun commit last adds reps flops =
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   876
          (if debug andalso auto_level = 0 then
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   877
             Output.urgent_message "Committing..."
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   878
           else
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   879
             ();
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   880
           mash_map ctxt (do_commit (rev adds) reps flops);
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   881
           if not last andalso auto_level = 0 then
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   882
             let val num_proofs = length adds + length reps in
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   883
               "Learned " ^ string_of_int num_proofs ^ " " ^
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   884
               (if atp then "ATP" else "Isar") ^ " proof" ^
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   885
               plural_s num_proofs ^ " in the last " ^
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   886
               string_from_time commit_timeout ^ "."
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   887
               |> Output.urgent_message
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   888
             end
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   889
           else
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   890
             ())
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   891
        fun learn_new_fact _ (accum as (_, (_, _, _, true))) = accum
48439
67a6bcbd3587 removed MaSh junk arising from primrec definitions
blanchet
parents: 48438
diff changeset
   892
          | learn_new_fact ((_, stature as (_, status)), th)
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   893
                           (adds, (parents, n, next_commit, _)) =
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
   894
            let
48378
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48377
diff changeset
   895
              val name = nickname_of th
48332
271a4a6af734 optimize parent computation in MaSh + remove temporary files
blanchet
parents: 48330
diff changeset
   896
              val feats =
48385
2779dea0b1e0 added locality as a MaSh feature
blanchet
parents: 48384
diff changeset
   897
                features_of ctxt prover (theory_of_thm th) stature [prop_of th]
48439
67a6bcbd3587 removed MaSh junk arising from primrec definitions
blanchet
parents: 48438
diff changeset
   898
              val deps = deps_of status th |> these
48394
82fc8c956cdc fixed various issues with MaSh's file handling + tune output + generate local facts again + handle nameless facts gracefully
blanchet
parents: 48392
diff changeset
   899
              val n = n |> not (null deps) ? Integer.add 1
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   900
              val adds = (name, parents, feats, deps) :: adds
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   901
              val (adds, next_commit) =
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   902
                if Time.> (Timer.checkRealTimer timer, next_commit) then
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   903
                  (commit false adds [] []; ([], next_commit_time ()))
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
   904
                else
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   905
                  (adds, next_commit)
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   906
              val timed_out = Time.> (Timer.checkRealTimer timer, learn_timeout)
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   907
            in (adds, ([name], n, next_commit, timed_out)) end
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   908
        val n =
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   909
          if null new_facts then
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   910
            0
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   911
          else
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   912
            let
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   913
              val last_th = new_facts |> List.last |> snd
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   914
              (* crude approximation *)
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   915
              val ancestors =
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   916
                old_facts
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   917
                |> filter (fn (_, th) => thm_ord (th, last_th) <> GREATER)
48407
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
   918
              val parents = maximal_in_graph fact_G ancestors
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   919
              val (adds, (_, n, _, _)) =
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   920
                ([], (parents, 0, next_commit_time (), false))
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   921
                |> fold learn_new_fact new_facts
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   922
            in commit true adds [] []; n end
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   923
        fun relearn_old_fact _ (accum as (_, (_, _, true))) = accum
48439
67a6bcbd3587 removed MaSh junk arising from primrec definitions
blanchet
parents: 48438
diff changeset
   924
          | relearn_old_fact ((_, (_, status)), th)
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   925
                             ((reps, flops), (n, next_commit, _)) =
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   926
            let
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   927
              val name = nickname_of th
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   928
              val (n, reps, flops) =
48439
67a6bcbd3587 removed MaSh junk arising from primrec definitions
blanchet
parents: 48438
diff changeset
   929
                case deps_of status th of
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   930
                  SOME deps => (n + 1, (name, deps) :: reps, flops)
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   931
                | NONE => (n, reps, name :: flops)
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   932
              val (reps, flops, next_commit) =
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   933
                if Time.> (Timer.checkRealTimer timer, next_commit) then
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   934
                  (commit false [] reps flops; ([], [], next_commit_time ()))
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   935
                else
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   936
                  (reps, flops, next_commit)
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   937
              val timed_out = Time.> (Timer.checkRealTimer timer, learn_timeout)
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   938
            in ((reps, flops), (n, next_commit, timed_out)) end
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   939
        val n =
48433
9e9b6e363859 don't relearn old facts in Isar mode
blanchet
parents: 48408
diff changeset
   940
          if not atp orelse null old_facts then
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   941
            n
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   942
          else
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   943
            let
48668
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   944
              val max_isar = 1000 * max_dependencies
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   945
              fun kind_of_proof th =
48668
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   946
                try (Graph.get_node fact_G) (nickname_of th)
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   947
                |> the_default Isar_Proof
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   948
              fun priority_of (_, th) =
48668
5d63c23b4042 remember which MaSh proofs were found using ATPs
blanchet
parents: 48667
diff changeset
   949
                random_range 0 max_isar
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   950
                + (case kind_of_proof th of
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   951
                     Isar_Proof => 0
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   952
                   | ATP_Proof => 2 * max_isar
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   953
                   | Isar_Proof_wegen_ATP_Flop => max_isar)
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   954
                - 500 * (th |> isar_dependencies_of all_names
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   955
                            |> Option.map length
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   956
                            |> the_default max_dependencies)
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   957
              val old_facts =
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   958
                old_facts |> map (`priority_of)
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   959
                          |> sort (int_ord o pairself fst)
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   960
                          |> map snd
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   961
              val ((reps, flops), (n, _, _)) =
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   962
                (([], []), (n, next_commit_time (), false))
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   963
                |> fold relearn_old_fact old_facts
48669
cdcdb0547f29 remember ATP flops to avoid repeating them too quickly
blanchet
parents: 48668
diff changeset
   964
            in commit true [] reps flops; n end
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
   965
      in
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   966
        if verbose orelse auto_level < 2 then
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   967
          "Learned " ^ string_of_int n ^ " nontrivial " ^
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   968
          (if atp then "ATP" else "Isar") ^ " proof" ^ plural_s n ^
48319
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
   969
          (if verbose then
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
   970
             " in " ^ string_from_time (Timer.checkRealTimer timer)
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
   971
           else
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
   972
             "") ^ "."
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
   973
        else
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
   974
          ""
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
   975
      end
48308
89674e5a4d35 make tracing an option
blanchet
parents: 48306
diff changeset
   976
  end
48304
50e64af9c829 more work on MaSh
blanchet
parents: 48303
diff changeset
   977
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   978
fun mash_learn ctxt (params as {provers, timeout, ...}) fact_override chained
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   979
               atp =
48316
252f45c04042 drastic overhaul of MaSh data structures + fixed a few performance issues
blanchet
parents: 48315
diff changeset
   980
  let
48396
dd82d190c2af name tuning
blanchet
parents: 48395
diff changeset
   981
    val css = Sledgehammer_Fact.clasimpset_rule_table_of ctxt
48395
85a7fb65507a learning should honor the fact override and the chained facts
blanchet
parents: 48394
diff changeset
   982
    val ctxt = ctxt |> Config.put instantiate_inducts false
85a7fb65507a learning should honor the fact override and the chained facts
blanchet
parents: 48394
diff changeset
   983
    val facts =
48396
dd82d190c2af name tuning
blanchet
parents: 48395
diff changeset
   984
      nearly_all_facts ctxt false fact_override Symtab.empty css chained []
dd82d190c2af name tuning
blanchet
parents: 48395
diff changeset
   985
                       @{prop True}
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   986
    val num_facts = length facts
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   987
    val prover = hd provers
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   988
    fun learn auto_level atp =
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   989
      mash_learn_facts ctxt params prover auto_level atp infinite_timeout facts
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
   990
      |> Output.urgent_message
48316
252f45c04042 drastic overhaul of MaSh data structures + fixed a few performance issues
blanchet
parents: 48315
diff changeset
   991
  in
50340
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
   992
    if atp then
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
   993
      ("MaShing through " ^ string_of_int num_facts ^ " fact" ^
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
   994
       plural_s num_facts ^ " for ATP proofs (" ^ quote prover ^ " timeout: " ^
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
   995
       string_from_time timeout ^ ").\n\nCollecting Isar proofs first..."
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
   996
       |> Output.urgent_message;
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
   997
       learn 1 false;
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
   998
       "Now collecting ATP proofs. This may take several hours. You can \
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
   999
       \safely stop the learning process at any point."
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
  1000
       |> Output.urgent_message;
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
  1001
       learn 0 true)
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
  1002
    else
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
  1003
      ("MaShing through " ^ string_of_int num_facts ^ " fact" ^
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
  1004
       plural_s num_facts ^ " for Isar proofs..."
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
  1005
       |> Output.urgent_message;
72519bf5f135 simplify MaSh term patterns + add missing global facts if there aren't too many
blanchet
parents: 50339
diff changeset
  1006
       learn 0 false)
48316
252f45c04042 drastic overhaul of MaSh data structures + fixed a few performance issues
blanchet
parents: 48315
diff changeset
  1007
  end
48249
2bd242c56c90 dummy implementation
blanchet
parents: 48248
diff changeset
  1008
50311
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
  1009
fun is_mash_enabled () = (getenv "MASH" = "yes")
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
  1010
fun mash_can_suggest_facts ctxt = not (Graph.is_empty (#fact_G (mash_get ctxt)))
c9d7ccd090e1 tuned order of functions
blanchet
parents: 50310
diff changeset
  1011
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
  1012
(* Generate more suggestions than requested, because some might be thrown out
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
  1013
   later for various reasons. *)
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
  1014
fun generous_max_facts max_facts = max_facts + Int.min (50, max_facts)
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
  1015
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
  1016
(* The threshold should be large enough so that MaSh doesn't kick in for Auto
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
  1017
   Sledgehammer and Try. *)
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
  1018
val min_secs_for_learning = 15
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
  1019
48321
c552d7f1720b learn from minimized ATP proofs
blanchet
parents: 48320
diff changeset
  1020
fun relevant_facts ctxt (params as {learn, fact_filter, timeout, ...}) prover
c552d7f1720b learn from minimized ATP proofs
blanchet
parents: 48320
diff changeset
  1021
        max_facts ({add, only, ...} : fact_override) hyp_ts concl_t facts =
48314
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
  1022
  if not (subset (op =) (the_list fact_filter, fact_filters)) then
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
  1023
    error ("Unknown fact filter: " ^ quote (the fact_filter) ^ ".")
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
  1024
  else if only then
48289
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48288
diff changeset
  1025
    facts
48321
c552d7f1720b learn from minimized ATP proofs
blanchet
parents: 48320
diff changeset
  1026
  else if max_facts <= 0 orelse null facts then
48288
255c6e1fd505 rationalize relevance filter, slowing moving code from Iter to MaSh
blanchet
parents: 48251
diff changeset
  1027
    []
255c6e1fd505 rationalize relevance filter, slowing moving code from Iter to MaSh
blanchet
parents: 48251
diff changeset
  1028
  else
255c6e1fd505 rationalize relevance filter, slowing moving code from Iter to MaSh
blanchet
parents: 48251
diff changeset
  1029
    let
48327
568b3193e53e don't include hidden facts in relevance filter + tweak MaSh learning
blanchet
parents: 48326
diff changeset
  1030
      fun maybe_learn () =
48384
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
  1031
        if learn andalso not (Async_Manager.has_running_threads MaShN) andalso
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
  1032
           Time.toSeconds timeout >= min_secs_for_learning then
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
  1033
          let val timeout = time_mult learn_timeout_slack timeout in
83dc102041e6 learn on explicit "min" command but do the learning in a thread, since it may take a couple of seconds
blanchet
parents: 48383
diff changeset
  1034
            launch_thread timeout
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48403
diff changeset
  1035
                (fn () => (true, mash_learn_facts ctxt params prover 2 false
48392
ca998fa08cd9 added "learn_from_atp" command to MaSh, for patient users
blanchet
parents: 48390
diff changeset
  1036
                                                  timeout facts))
48319
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
  1037
          end
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
  1038
        else
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
  1039
          ()
48314
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
  1040
      val fact_filter =
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
  1041
        case fact_filter of
48379
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
  1042
          SOME ff => (() |> ff <> mepoN ? maybe_learn; ff)
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
  1043
        | NONE =>
48407
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
  1044
          if is_smt_prover ctxt prover then
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
  1045
            mepoN
50229
78f3be5e51d2 simplify code slightly
blanchet
parents: 50220
diff changeset
  1046
          else if is_mash_enabled () then
48407
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
  1047
            (maybe_learn ();
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
  1048
             if mash_can_suggest_facts ctxt then meshN else mepoN)
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
  1049
          else
47fe0ca12fc2 faster maximal node computation
blanchet
parents: 48406
diff changeset
  1050
            mepoN
48288
255c6e1fd505 rationalize relevance filter, slowing moving code from Iter to MaSh
blanchet
parents: 48251
diff changeset
  1051
      val add_ths = Attrib.eval_thms ctxt add
48292
7fcee834c7f5 more code rationalization in relevance filter
blanchet
parents: 48289
diff changeset
  1052
      fun prepend_facts ths accepts =
48288
255c6e1fd505 rationalize relevance filter, slowing moving code from Iter to MaSh
blanchet
parents: 48251
diff changeset
  1053
        ((facts |> filter (member Thm.eq_thm_prop ths o snd)) @
48292
7fcee834c7f5 more code rationalization in relevance filter
blanchet
parents: 48289
diff changeset
  1054
         (accepts |> filter_out (member Thm.eq_thm_prop ths o snd)))
48293
914ca0827804 renamed Sledgehammer options
blanchet
parents: 48292
diff changeset
  1055
        |> take max_facts
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
  1056
      fun mepo () =
50382
blanchet
parents: 50359
diff changeset
  1057
        mepo_suggested_facts ctxt params prover max_facts NONE hyp_ts concl_t
blanchet
parents: 50359
diff changeset
  1058
                             facts
blanchet
parents: 50359
diff changeset
  1059
        |> weight_mepo_facts
48314
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
  1060
      fun mash () =
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
  1061
        mash_suggested_facts ctxt params prover (generous_max_facts max_facts)
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
  1062
            hyp_ts concl_t facts
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
  1063
        |> weight_mash_facts
48314
ee33ba3c0e05 added option to control which fact filter is used
blanchet
parents: 48313
diff changeset
  1064
      val mess =
50383
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
  1065
        [] |> (if fact_filter <> mashN then cons (0.5, (mepo (), [])) else I)
4274b25ff4e7 take proximity into account for MaSh + fix a debilitating bug in feature generation
blanchet
parents: 50382
diff changeset
  1066
           |> (if fact_filter <> mepoN then cons (0.5, (mash (), [])) else I)
48288
255c6e1fd505 rationalize relevance filter, slowing moving code from Iter to MaSh
blanchet
parents: 48251
diff changeset
  1067
    in
48313
0faafdffa662 mesh facts by taking into consideration whether a fact is known to MeSh
blanchet
parents: 48312
diff changeset
  1068
      mesh_facts max_facts mess
48288
255c6e1fd505 rationalize relevance filter, slowing moving code from Iter to MaSh
blanchet
parents: 48251
diff changeset
  1069
      |> not (null add_ths) ? prepend_facts add_ths
255c6e1fd505 rationalize relevance filter, slowing moving code from Iter to MaSh
blanchet
parents: 48251
diff changeset
  1070
    end
255c6e1fd505 rationalize relevance filter, slowing moving code from Iter to MaSh
blanchet
parents: 48251
diff changeset
  1071
48319
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
  1072
fun kill_learners () = Async_Manager.kill_threads MaShN "learner"
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
  1073
fun running_learners () = Async_Manager.running_threads MaShN "learner"
340187063d84 use async manager to manage MaSh learners to make sure they get killed cleanly
blanchet
parents: 48318
diff changeset
  1074
48248
b6eb45a52c28 split relevance filter code into three files
blanchet
parents:
diff changeset
  1075
end;