run-eval

#!/bin/bash
# run-eval: run fasttext ten times and evaluate
# usage: run-eval
# 20171002 erikt(at)xs4all.nl

TRAIN=dutch-2012.train-dev.txt
TEST=dutch-2012.dev.txt
PYTHONDIR=$HOME/projects/online-behaviour/machine-learning
TMPFILE=run-eval.$$.$RANDOM
OUTFILE=run-eval.out.$$.$RANDOM
TMPTRAIN=$TMPFILE.train
TMPTEST=$TMPFILE.test
DIM=300
MINCOUNT=5

function prepare {
   sed 's/ ID=[^ ]* / /' | sed 's/ SENDER=[^ ]* / /' | sed 's/ RAWTEXT .*$//'
}
prepare < $TRAIN > $TMPTRAIN
prepare < $TEST > $TMPTEST

(
   echo "# TRAIN=$TRAIN TEST=$TEST DIM=$DIM MINCOUNT=$MINCOUNT"
   for EXP in a b c d e f g h i j
   do
      VECTORS=vectors-train-dev-$EXP.vec
      ../fasttext supervised -input $TMPTRAIN -output $TMPFILE \
         -dim $DIM -minCount $MINCOUNT -pretrainedVectors $VECTORS \
	  >/dev/null 2>/dev/null
      ../fasttext predict $TMPFILE.bin $TMPTEST |\
         paste -d ' ' - $TMPTEST | cut -d' ' -f1,2 | $PYTHONDIR/eval.py |\
         head -1 | rev | sed 's/^ *//' | cut -d' ' -f1 | rev
   done) | tee $OUTFILE | $PYTHONDIR/average.py > $TMPFILE.average
sed 's/^/# /' $TMPFILE.average >> $OUTFILE
cat $TMPFILE.average | sed "s/$/ $OUTFILE/"

rm -f $TMPFILE.*
exit 0