blanchet [Tue, 24 Aug 2010 14:36:29 +0200] rev 38690
make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet [Mon, 23 Aug 2010 23:32:11 +0200] rev 38689
cosmetics
blanchet [Mon, 23 Aug 2010 23:24:24 +0200] rev 38688
optimize fact filter some more
blanchet [Mon, 23 Aug 2010 23:00:57 +0200] rev 38687
cache previous iteration's weights, and keep track of what's dirty and what's clean;
this speeds up the relevance filter significantly
blanchet [Mon, 23 Aug 2010 21:55:52 +0200] rev 38686
modified relevance filter
blanchet [Mon, 23 Aug 2010 21:11:30 +0200] rev 38685
adjust fudge factors in the light of experiments
blanchet [Mon, 23 Aug 2010 18:53:11 +0200] rev 38684
invert semantics of "relevance_convergence", to make it more intuitive
blanchet [Mon, 23 Aug 2010 18:39:12 +0200] rev 38683
if no facts were selected on first iteration, try again with a lower threshold
blanchet [Mon, 23 Aug 2010 18:25:49 +0200] rev 38682
weed out junk in relevance filter
blanchet [Mon, 23 Aug 2010 18:04:31 +0200] rev 38681
no need for "eq" facts -- good old "=" is good enough for us