pattern_mm.sh

   1 #!/bin/sh
   2 # pattern_mm - Harvest patterns from game collection and compute feature strengths
   3 #
   4 # This is a "frontend" for the MM tool by Remi Coulom, that is assumed to be
   5 # unpacked and compiled in the mm/ subdirectory - get it at
   6 #
   7 #       http://remi.coulom.free.fr/Amsterdam2007/mm.tar.bz2
   8 #
   9 # It will scan a given SGF collection, collect patterns, and use the MM tool
  10 # to compute the relative strength of various features. The output will be
  11 #
  12 #       * patterns.gamma: Gamma values of pattern features
  13 #       * patterns.gammaf: Gamma values of pattern features for MC simulations
  14 #
  15 # If you haven't done so yet, you should first run ./pattern_spatial_gen.sh
  16 # (probably in the competition scan mode) to initialize the spatial patterns
  17 # dictionary for the collection.
  18 #
  19 # If you run this on hundreds of games, be sure you are doing it on local
  20 # filesystem, with some free memory (and few GB of free disk on both
  21 # local fs and in /tmp for temporary data), and armed by a lot of patience
  22 # - it can take long time (minutes, tens of minutes...).
  23
  24
  25 if [ -z "$mm_file" ]; then
  26         mm_file=patterns.gamma mm_par= ./pattern_mm.sh "$@"
  27         mm_file=patterns.gammaf mm_par=,matchfast ./pattern_mm.sh "$@"
  28         exit
  29 fi
  30
  31 echo "Gathering patterns and feeding MM..."
  32 (for i in "$@"; do ./sgf2gtp.pl $i; done) |
  33         ./zzgo -d 0 -e patternscan competition,mm$mm_par |
  34         sed -e '/^= $/d; s/^= /#\n/' | grep -v '^$' |
  35         mm/mm >/tmp/patterns.mm
  36 ls -l /tmp/patterns.mm
  37
  38 # There must not be pipeline here, because of aux patterns.fdict file!
  39
  40 echo "Associating gamma values..."
  41 cat /tmp/patterns.mm | sed 's/  */ /; s/^ //;' | join -o 2.2,1.2 /tmp/patterns.mm patterns.fdict | sed 's/^s\.[0-9]*:/s:/' >$mm_file
  42
  43 rm -f /tmp/patterns.mm
  44 echo "Product:"
  45 ls -l $mm_file
  46 echo "Leaving behind for analysis:"
  47 ls -l patterns.fdict mm-with-freq.dat