src/HOL/TPTP/mash_eval.ML
author blanchet
Thu, 26 Jul 2012 10:48:03 +0200
changeset 48530 d443166f9520
parent 48438 3e45c98fe127
child 48615 d5c9917ff5b6
permissions -rw-r--r--
repaired accessibility chains generated by MaSh exporter + tuned one function out
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
48285
902ab51dd12a renaming
blanchet
parents: 48284
diff changeset
     1
(*  Title:      HOL/TPTP/mash_eval.ML
48234
06216c789ac9 moved MaSh into own files
blanchet
parents:
diff changeset
     2
    Author:     Jasmin Blanchette, TU Muenchen
06216c789ac9 moved MaSh into own files
blanchet
parents:
diff changeset
     3
    Copyright   2012
06216c789ac9 moved MaSh into own files
blanchet
parents:
diff changeset
     4
48285
902ab51dd12a renaming
blanchet
parents: 48284
diff changeset
     5
Evaluate proof suggestions from MaSh (Machine-learning for Sledgehammer).
48234
06216c789ac9 moved MaSh into own files
blanchet
parents:
diff changeset
     6
*)
06216c789ac9 moved MaSh into own files
blanchet
parents:
diff changeset
     7
48285
902ab51dd12a renaming
blanchet
parents: 48284
diff changeset
     8
signature MASH_EVAL =
48234
06216c789ac9 moved MaSh into own files
blanchet
parents:
diff changeset
     9
sig
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48250
diff changeset
    10
  type params = Sledgehammer_Provers.params
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48250
diff changeset
    11
48285
902ab51dd12a renaming
blanchet
parents: 48284
diff changeset
    12
  val evaluate_mash_suggestions :
48251
6cdcfbddc077 moved most of MaSh exporter code to Sledgehammer
blanchet
parents: 48250
diff changeset
    13
    Proof.context -> params -> theory -> string -> unit
48234
06216c789ac9 moved MaSh into own files
blanchet
parents:
diff changeset
    14
end;
06216c789ac9 moved MaSh into own files
blanchet
parents:
diff changeset
    15
48285
902ab51dd12a renaming
blanchet
parents: 48284
diff changeset
    16
structure MaSh_Eval : MASH_EVAL =
48234
06216c789ac9 moved MaSh into own files
blanchet
parents:
diff changeset
    17
struct
48235
40655464a93b MaSh evaluation driver
blanchet
parents: 48234
diff changeset
    18
48315
82d6e46c673f fixed order of accessibles + other tweaks to MaSh
blanchet
parents: 48313
diff changeset
    19
open Sledgehammer_Fact
48381
1b7d798460bb renamed ML structures
blanchet
parents: 48379
diff changeset
    20
open Sledgehammer_MaSh
48240
6a8d18798161 rationalized output
blanchet
parents: 48239
diff changeset
    21
48379
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
    22
val MePoN = "MePo"
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
    23
val MaShN = "MaSh"
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
    24
val MeshN = "Mesh"
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
    25
val IsarN = "Isar"
48241
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
    26
48293
914ca0827804 renamed Sledgehammer options
blanchet
parents: 48292
diff changeset
    27
val max_facts_slack = 2
48235
40655464a93b MaSh evaluation driver
blanchet
parents: 48234
diff changeset
    28
48438
3e45c98fe127 distinguish between recursive and nonrecursive definitions + clean up typedef dependencies in MaSh
blanchet
parents: 48406
diff changeset
    29
val all_names = map (rpair () o nickname_of) #> Symtab.make
48316
252f45c04042 drastic overhaul of MaSh data structures + fixed a few performance issues
blanchet
parents: 48315
diff changeset
    30
48285
902ab51dd12a renaming
blanchet
parents: 48284
diff changeset
    31
fun evaluate_mash_suggestions ctxt params thy file_name =
48235
40655464a93b MaSh evaluation driver
blanchet
parents: 48234
diff changeset
    32
  let
48293
914ca0827804 renamed Sledgehammer options
blanchet
parents: 48292
diff changeset
    33
    val {provers, max_facts, slice, type_enc, lam_trans, timeout, ...} =
48239
0016290f904c generate Meng--Paulson facts for evaluation purposes
blanchet
parents: 48236
diff changeset
    34
      Sledgehammer_Isar.default_params ctxt []
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
    35
    val prover = hd provers
48313
0faafdffa662 mesh facts by taking into consideration whether a fact is known to MeSh
blanchet
parents: 48312
diff changeset
    36
    val slack_max_facts = max_facts_slack * the max_facts
48235
40655464a93b MaSh evaluation driver
blanchet
parents: 48234
diff changeset
    37
    val path = file_name |> Path.explode
40655464a93b MaSh evaluation driver
blanchet
parents: 48234
diff changeset
    38
    val lines = path |> File.read_lines
48530
d443166f9520 repaired accessibility chains generated by MaSh exporter + tuned one function out
blanchet
parents: 48438
diff changeset
    39
    val css = Sledgehammer_Fact.clasimpset_rule_table_of ctxt
d443166f9520 repaired accessibility chains generated by MaSh exporter + tuned one function out
blanchet
parents: 48438
diff changeset
    40
    val facts =
d443166f9520 repaired accessibility chains generated by MaSh exporter + tuned one function out
blanchet
parents: 48438
diff changeset
    41
      all_facts (Proof_Context.init_global thy) false Symtab.empty [] [] css
48324
3ee5b5589402 speed up tautology/metaness check
blanchet
parents: 48321
diff changeset
    42
    val all_names = all_names (facts |> map snd)
48379
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
    43
    val mepo_ok = Unsynchronized.ref 0
48241
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
    44
    val mash_ok = Unsynchronized.ref 0
48298
f5b160f9859e more work on MaSh
blanchet
parents: 48293
diff changeset
    45
    val mesh_ok = Unsynchronized.ref 0
48300
9910021c80a7 tweak output
blanchet
parents: 48299
diff changeset
    46
    val isar_ok = Unsynchronized.ref 0
48289
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    47
    fun with_index facts s = (find_index (curry (op =) s) facts + 1, s)
48235
40655464a93b MaSh evaluation driver
blanchet
parents: 48234
diff changeset
    48
    fun index_string (j, s) = s ^ "@" ^ string_of_int j
48240
6a8d18798161 rationalized output
blanchet
parents: 48239
diff changeset
    49
    fun str_of_res label facts {outcome, run_time, used_facts, ...} =
48289
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    50
      let val facts = facts |> map (fn ((name, _), _) => name ()) in
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    51
        "  " ^ label ^ ": " ^
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    52
        (if is_none outcome then
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    53
           "Success (" ^ ATP_Util.string_from_time run_time ^ "): " ^
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    54
           (used_facts |> map (with_index facts o fst)
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    55
                       |> sort (int_ord o pairself fst)
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    56
                       |> map index_string
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    57
                       |> space_implode " ") ^
48293
914ca0827804 renamed Sledgehammer options
blanchet
parents: 48292
diff changeset
    58
           (if length facts < the max_facts then
48289
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    59
              " (of " ^ string_of_int (length facts) ^ ")"
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    60
            else
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    61
              "")
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    62
         else
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    63
           "Failure: " ^
48293
914ca0827804 renamed Sledgehammer options
blanchet
parents: 48292
diff changeset
    64
           (facts |> take (the max_facts) |> tag_list 1
48289
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    65
                  |> map index_string
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    66
                  |> space_implode " "))
6b65f1ad0e4b systematize lazy names in relevance filter
blanchet
parents: 48285
diff changeset
    67
      end
48311
3c4e10606567 implemented MaSh QUERY operation
blanchet
parents: 48307
diff changeset
    68
    fun solve_goal (j, line) =
48235
40655464a93b MaSh evaluation driver
blanchet
parents: 48234
diff changeset
    69
      let
48311
3c4e10606567 implemented MaSh QUERY operation
blanchet
parents: 48307
diff changeset
    70
        val (name, suggs) = extract_query line
48235
40655464a93b MaSh evaluation driver
blanchet
parents: 48234
diff changeset
    71
        val th =
48378
9e96486d53ad handle local facts smoothly in MaSh
blanchet
parents: 48333
diff changeset
    72
          case find_first (fn (_, th) => nickname_of th = name) facts of
48235
40655464a93b MaSh evaluation driver
blanchet
parents: 48234
diff changeset
    73
            SOME (_, th) => th
48530
d443166f9520 repaired accessibility chains generated by MaSh exporter + tuned one function out
blanchet
parents: 48438
diff changeset
    74
          | NONE => error ("No fact called \"" ^ name ^ "\".")
48250
1065c307fafe further ML structure split to permit finer-grained loading/reordering (problem to solve: MaSh needs most of Sledgehammer)
blanchet
parents: 48245
diff changeset
    75
        val goal = goal_of_thm thy th
48245
854a47677335 generate ATP dependencies
blanchet
parents: 48241
diff changeset
    76
        val (_, hyp_ts, concl_t) = ATP_Util.strip_subgoal ctxt goal 1
48404
0a261b4aa093 relearn ATP proofs
blanchet
parents: 48394
diff changeset
    77
        val isar_deps = isar_dependencies_of all_names th |> these
48235
40655464a93b MaSh evaluation driver
blanchet
parents: 48234
diff changeset
    78
        val facts = facts |> filter (fn (_, th') => thm_ord (th', th) = LESS)
48379
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
    79
        val mepo_facts =
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    80
          Sledgehammer_MePo.mepo_suggested_facts ctxt params prover
48381
1b7d798460bb renamed ML structures
blanchet
parents: 48379
diff changeset
    81
              slack_max_facts NONE hyp_ts concl_t facts
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    82
          |> Sledgehammer_MePo.weight_mepo_facts
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    83
        val mash_facts = suggested_facts suggs facts
48379
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
    84
        val mess = [(mepo_facts, []), (mash_facts, [])]
48313
0faafdffa662 mesh facts by taking into consideration whether a fact is known to MeSh
blanchet
parents: 48312
diff changeset
    85
        val mesh_facts = mesh_facts slack_max_facts mess
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    86
        val isar_facts = suggested_facts (map (rpair 1.0) isar_deps) facts
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    87
        fun prove ok heading get facts =
48313
0faafdffa662 mesh facts by taking into consideration whether a fact is known to MeSh
blanchet
parents: 48312
diff changeset
    88
          let
0faafdffa662 mesh facts by taking into consideration whether a fact is known to MeSh
blanchet
parents: 48312
diff changeset
    89
            val facts =
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    90
              facts |> map get
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    91
                    |> Sledgehammer_Fact.maybe_instantiate_inducts ctxt hyp_ts
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    92
                                                                   concl_t
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    93
                    |> take (the max_facts)
48321
c552d7f1720b learn from minimized ATP proofs
blanchet
parents: 48320
diff changeset
    94
            val res as {outcome, ...} =
c552d7f1720b learn from minimized ATP proofs
blanchet
parents: 48320
diff changeset
    95
              run_prover_for_mash ctxt params prover facts goal
48313
0faafdffa662 mesh facts by taking into consideration whether a fact is known to MeSh
blanchet
parents: 48312
diff changeset
    96
            val _ = if is_none outcome then ok := !ok + 1 else ()
0faafdffa662 mesh facts by taking into consideration whether a fact is known to MeSh
blanchet
parents: 48312
diff changeset
    97
          in str_of_res heading facts res end
48406
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    98
        val mepo_s = prove mepo_ok MePoN fst mepo_facts
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
    99
        val mash_s = prove mash_ok MaShN fst mash_facts
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   100
        val mesh_s = prove mesh_ok MeshN I mesh_facts
b002cc16aa99 honor suggested MaSh weights
blanchet
parents: 48404
diff changeset
   101
        val isar_s = prove isar_ok IsarN fst isar_facts
48235
40655464a93b MaSh evaluation driver
blanchet
parents: 48234
diff changeset
   102
      in
48379
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
   103
        ["Goal " ^ string_of_int j ^ ": " ^ name, mepo_s, mash_s, mesh_s,
48300
9910021c80a7 tweak output
blanchet
parents: 48299
diff changeset
   104
         isar_s]
48241
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
   105
        |> cat_lines |> tracing
48235
40655464a93b MaSh evaluation driver
blanchet
parents: 48234
diff changeset
   106
      end
48241
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
   107
    fun total_of heading ok n =
48379
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
   108
      "  " ^ heading ^ ": " ^ string_of_int (!ok) ^ " (" ^
48241
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
   109
      Real.fmt (StringCvt.FIX (SOME 1))
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
   110
               (100.0 * Real.fromInt (!ok) / Real.fromInt n) ^ "%)"
48250
1065c307fafe further ML structure split to permit finer-grained loading/reordering (problem to solve: MaSh needs most of Sledgehammer)
blanchet
parents: 48245
diff changeset
   111
    val inst_inducts = Config.get ctxt Sledgehammer_Fact.instantiate_inducts
48245
854a47677335 generate ATP dependencies
blanchet
parents: 48241
diff changeset
   112
    val options =
48318
325c8fd0d762 more consolidation of MaSh code
blanchet
parents: 48316
diff changeset
   113
      [prover, string_of_int (the max_facts) ^ " facts",
48241
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
   114
       "slice" |> not slice ? prefix "dont_", the_default "smart" type_enc,
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
   115
       the_default "smart" lam_trans, ATP_Util.string_from_time timeout,
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
   116
       "instantiate_inducts" |> not inst_inducts ? prefix "dont_"]
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
   117
    val n = length lines
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
   118
  in
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
   119
    tracing " * * *";
48245
854a47677335 generate ATP dependencies
blanchet
parents: 48241
diff changeset
   120
    tracing ("Options: " ^ commas options);
48311
3c4e10606567 implemented MaSh QUERY operation
blanchet
parents: 48307
diff changeset
   121
    List.app solve_goal (tag_list 1 lines);
48241
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
   122
    ["Successes (of " ^ string_of_int n ^ " goals)",
48379
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
   123
     total_of MePoN mepo_ok n,
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
   124
     total_of MaShN mash_ok n,
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
   125
     total_of MeshN mesh_ok n,
2b5ad61e2ccc renamed "iter" fact filter to "MePo" (Meng--Paulson)
blanchet
parents: 48378
diff changeset
   126
     total_of IsarN isar_ok n]
48241
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
   127
    |> cat_lines |> tracing
688f1172d523 nicer output
blanchet
parents: 48240
diff changeset
   128
  end
48235
40655464a93b MaSh evaluation driver
blanchet
parents: 48234
diff changeset
   129
48234
06216c789ac9 moved MaSh into own files
blanchet
parents:
diff changeset
   130
end;