Added argument pretrained= to train.sh, to chose which column to apply Glove to

UD_any/data/Makefile

@@ -69,7 +69,7 @@ texts_lines:
 	./getRawText.py $(CONLL2LINES) $(TRAIN_FILES) $(DEV_FILES) $(TEST_FILES)
 
 pretrain:
-	for col in FORM UPOS LEMMA FEATS DEPREL LETTERS ; do \
+	for col in $(PRETRAINED_COLS) ; do \
 	./pretrainEmbeddings.sh $(TRAIN_FILES) $$col 128 $$col.w2v 2> pretrain_log.err || ( cat pretrain_log.err && exit 1 ) ; \
 	done
 

UD_any/launchBatches.py

@@ -8,7 +8,7 @@ from launchSlurmArray import launchSlurmArray
 
 ###############################################################################
 def printUsageAndExit() :
-  print("USAGE : %s (train | eval) (gpu | cpu) batchesDescription.py nbHours jobName maxNbSimultaneousJobs"%sys.argv[0], file=sys.stderr)
+  print("USAGE : %s (train | eval) (gpu | cpu) nbCPU batchesDescription.py nbHours jobName maxNbSimultaneousJobs"%sys.argv[0], file=sys.stderr)
   exit(1)
 ###############################################################################
 
@@ -19,31 +19,32 @@ def prepareExperiment(lang, template, expName) :
 ###############################################################################
 
 ###############################################################################
-def addNamesAndCommandsTrain(names, commands, mode, expName, arguments, seed) :
+def addNamesAndCommandsTrain(names, commands, mode, expName, arguments, seed, pretrained) :
   names.append(expName)
   
-  commands.append("./train.sh {} bin/{} {} --silent --seed {}".format(mode, expName,arguments,seed))
+  commands.append("./train.sh {} bin/{} pretrained={} {} --silent --seed {}".format(mode, expName,pretrained,arguments,seed))
 ###############################################################################
 
 ###############################################################################
-def addNamesAndCommandsDecode(names, commands, mode, expName, arguments) :
+def addNamesAndCommandsDecode(names, commands, mode, expName, arguments, pretrained) :
   names.append(expName)
   
-  commands.append("./evaluate.sh {} bin/{} --silent {}".format(mode, expName, arguments))
+  commands.append("./evaluate.sh {} bin/{} pretrained={} --silent {}".format(mode, expName, pretrained,arguments))
 ###############################################################################
 
 ###############################################################################
 if __name__ == "__main__" :
 
-  if len(sys.argv) != 7 :
+  if len(sys.argv) != 8 :
     printUsageAndExit()
 
   mode = sys.argv[1]
   device = sys.argv[2]
-  batchesDescription = sys.argv[3]
-  nbHours = sys.argv[4]
-  name = sys.argv[5]
-  limit = sys.argv[6]
+  nbCPU = sys.argv[3]
+  batchesDescription = sys.argv[4]
+  nbHours = sys.argv[5]
+  name = sys.argv[6]
+  limit = sys.argv[7]
 
   if mode not in ["train","eval"] or device not in ["cpu","gpu"] :
     printUsageAndExit()
@@ -55,15 +56,16 @@ if __name__ == "__main__" :
 
   for lang in desc.langs :
     for xp in desc.templatesExperiments :
+      pretrained = xp['pretrained'] if "pretrained" in xp else ""
       for i in desc.repRange :
         xp['lang'] = lang
         xp['expName'] = xp['expName'].split('.')[0]+"."+lang+"."+str(i)
         if mode == "train" :
           prepareExperiment(xp['lang'],xp['template'],xp['expName'])
-          addNamesAndCommandsTrain(names, commands, xp['mode'],xp['expName'],xp['arguments'],seed=100+i)
+          addNamesAndCommandsTrain(names, commands, xp['mode'],xp['expName'],xp['arguments'],seed=100+i, pretrained=pretrained)
         else :
-          addNamesAndCommandsDecode(names, commands, xp['mode'],xp['expName'],xp['evalArguments'])
+          addNamesAndCommandsDecode(names, commands, xp['mode'],xp['expName'],xp['evalArguments'], pretrained=pretrained)
 
-  launchSlurmArray(names, commands, name, device, nbHours, limit)
+  launchSlurmArray(names, commands, name, device, nbHours, limit, nbCPU)
 ###############################################################################
 

UD_any/templates/tagger/machine.rm

@@ -4,9 +4,9 @@ Classifier : tagger
 	Transitions : {tagger,data/tagger.ts}
 	LossMultiplier : {}
 	Network type : Modular
-  Contextual : Window{-10 10} Columns{FORM} LSTM{1 1 0 1} In{64} Out{128} w2v{FORM,data/FORM.w2v} Targets{b.-2 b.-1 b.0 b.1 b.2}
+  Contextual : Window{-10 10} Columns{FORM} LSTM{1 1 0 1} In{128} Out{128} w2v{FORM,data/FORM.w2v} Targets{b.-2 b.-1 b.0 b.1 b.2}
   Context : Targets{b.-2 b.-1 b.0 b.1 b.2} Columns{EOS ID} LSTM{1 1 0 1} In{64} Out{64} w2v{} 
-  Context : Targets{b.-3 b.-2 b.-1} Columns{UPOS} LSTM{1 1 0 1} In{64} Out{64} w2v{UPOS,data/UPOS.w2v} 
+  Context : Targets{b.-3 b.-2 b.-1} Columns{UPOS} LSTM{1 1 0 1} In{64} Out{64} w2v{} 
   Focused : Column{prefix3:FORM} NbElem{3} Buffer{0} Stack{} LSTM{1 1 0 1} In{64} Out{64} w2v{}
   Focused : Column{suffix3:FORM} NbElem{3} Buffer{0} Stack{} LSTM{1 1 0 1} In{64} Out{64} w2v{}
 	InputDropout : 0.3

UD_any/train.sh

 #! /usr/bin/env bash
 
 function print_usage_and_exit {
-  >&2 echo "USAGE : (tsv | txt) expPath [arguments]"
+  >&2 echo "USAGE : (tsv | txt) expPath pretrained=col1,col2... [arguments]"
   exit 1
 }
 
 MODE=$1
 EXPPATH=$2
+PRETRAINED=$3
 
 >&2 echo "********************************************************************************"
 >&2 echo "Training : "$EXPPATH
@@ -23,6 +24,15 @@ then
   print_usage_and_exit
 fi
 
+if [ -z "$PRETRAINED" ];
+then
+  >&2 echo "ERROR : missing argument 3 (pretrained)"
+  print_usage_and_exit
+fi
+
+PRETRAINED=$(python3 -c 'import sys; print(" ".join(sys.argv[-1].split("=")[-1].split(",")))' $PRETRAINED)
+
+shift
 shift
 shift
 
@@ -38,7 +48,7 @@ then
 fi
 
 CURDIR=$(pwd)
-cd $EXPPATH"/"data && make -s clean && make $TARGET -s
+cd $EXPPATH"/"data && make -s clean && PRETRAINED_COLS=$PRETRAINED make $TARGET -s
 cd $CURDIR
 
 TRAIN=$EXPPATH"/data/train.conllu"

scripts/launchSlurmArray.py

 import os
 import subprocess
 
-def launchSlurmArray(names, commands, jobName, device, nbHours, limit) :
+def launchSlurmArray(names, commands, jobName, device, nbHours, limit, nbCPU) :
   commands = ["'%s'"%s for s in commands]
   names = ["'%s'"%s for s in names]
 
@@ -21,7 +21,7 @@ def launchSlurmArray(names, commands, jobName, device, nbHours, limit) :
 #SBATCH --error=%A_%a.err
 #SBATCH --open-mode=append
 #SBATCH --ntasks=1
-#SBATCH --cpus-per-task=4
+#SBATCH --cpus-per-task={}
 #SBATCH --partition=skylake
 #SBATCH -A b250
 #SBATCH --time={}:00:00
@@ -55,7 +55,7 @@ cat $tmpFile >> $newErr
 rm $tmpFile
 
 eval "${{commands[$SLURM_ARRAY_TASK_ID]}}"
-""".format(len(names)-1, limit, jobName, nbHours, " ".join(names), " ".join(commands)), file=sFile)
+""".format(len(names)-1, limit, jobName, nbCPU, nbHours, " ".join(names), " ".join(commands)), file=sFile)
     sFile.close()
   elif "jean-zay" in hostname :
     print("""#! /usr/bin/env bash
@@ -111,7 +111,7 @@ eval "${{commands[$SLURM_ARRAY_TASK_ID]}}"
 #SBATCH --error=%A_%a.err
 #SBATCH --open-mode=append
 #SBATCH --ntasks=1
-#SBATCH --cpus-per-task=4
+#SBATCH --cpus-per-task={}
 #SBATCH --partition={}
 #SBATCH --time={}:00:00
 
@@ -141,7 +141,7 @@ cat $tmpFile >> $newErr
 rm $tmpFile
 
 eval "${{commands[$SLURM_ARRAY_TASK_ID]}}"
-'''.format(len(names)-1, limit, jobName, "cpu" if device == "cpu" else "gpu\n#SBATCH --gres=gpu", nbHours, " ".join(names), commandList), file=sFile)
+'''.format(len(names)-1, limit, jobName, nbCPU, "cpu" if device == "cpu" else "gpu\n#SBATCH --gres=gpu", nbHours, " ".join(names), commandList), file=sFile)
     sFile.close()
   else :
     print("ERROR : Unknown hostname \'%s\'"%hostname)