New parameters : Nb of unaplayable actions picked before breaking the analysis

Train.py

@@ -16,7 +16,7 @@ import Config
 from conll18_ud_eval import load_conllu, evaluate
 
 ################################################################################
-def trainMode(debug, filename, type, transitionSet, strategy, modelDir, nbIter, batchSize, devFile, bootstrapInterval, incremental, rewardFunc, lr, gamma, probas, predicted, silent=False) :
+def trainMode(debug, filename, type, transitionSet, strategy, modelDir, nbIter, batchSize, devFile, bootstrapInterval, incremental, rewardFunc, lr, gamma, probas, countBreak, predicted, silent=False) :
   sentences = Config.readConllu(filename, predicted)
 
   if type == "oracle" :
@@ -24,7 +24,7 @@ def trainMode(debug, filename, type, transitionSet, strategy, modelDir, nbIter,
     return
 
   if type == "rl":
-    trainModelRl(debug, modelDir, filename, nbIter, batchSize, devFile, transitionSet, strategy, sentences, incremental, rewardFunc, lr, gamma, probas, predicted, silent)
+    trainModelRl(debug, modelDir, filename, nbIter, batchSize, devFile, transitionSet, strategy, sentences, incremental, rewardFunc, lr, gamma, probas, countBreak, predicted, silent)
     return
 
   print("ERROR : unknown type '%s'"%type, file=sys.stderr)
@@ -56,15 +56,15 @@ def extractExamples(debug, ts, strat, config, dicts, network, dynamic) :
         candidate = [[cand[0],cand[2]] for cand in scores if cand[1]][0][1]
         if debug :
           print(str(candidate), file=sys.stderr)
-        
+
       goldIndex = [str(trans) for trans in ts].index(str(candidateOracle))
       example = torch.cat([torch.LongTensor([goldIndex]), features])
       examples.append(example)
 
       moved = applyTransition(strat, config, candidate, None)
-  
+
     EOS.apply(config, strat)
-  
+
   return examples
 ################################################################################
 
@@ -149,7 +149,7 @@ def trainModelOracle(debug, modelDir, filename, nbEpochs, batchSize, devFile, tr
 ################################################################################
 
 ################################################################################
-def trainModelRl(debug, modelDir, filename, nbIter, batchSize, devFile, transitionSet, strategy, sentencesOriginal, incremental, rewardFunc, lr, gamma, probas, predicted, silent=False) :
+def trainModelRl(debug, modelDir, filename, nbIter, batchSize, devFile, transitionSet, strategy, sentencesOriginal, incremental, rewardFunc, lr, gamma, probas, countBreak, predicted, silent=False) :
 
   memory = None
   dicts = Dicts()
@@ -189,6 +189,8 @@ def trainModelRl(debug, modelDir, filename, nbIter, batchSize, devFile, transiti
       sentence.moveWordIndex(0)
       state = policy_net.extractFeatures(dicts, sentence).to(getDevice())
 
+      count = 0
+
       while True :
         missingLinks = getMissingLinks(sentence)
         if debug :
@@ -206,11 +208,14 @@ def trainModelRl(debug, modelDir, filename, nbIter, batchSize, devFile, transiti
         reward_ = rewarding(appliable, sentence, action, missingLinks, rewardFunc)
         reward = torch.FloatTensor([reward_]).to(getDevice())
 
-        newState = None
+        #newState = None
         if appliable :
           applyTransition(strategy, sentence, action, reward_)
           newState = policy_net.extractFeatures(dicts, sentence).to(getDevice())
 
+        else:
+          count+=1
+
         if memory is None :
           memory = ReplayMemory(5000, state.numel())
         memory.push(state, torch.LongTensor([transitionSet.index(action)]).to(getDevice()), newState, reward)
@@ -223,7 +228,7 @@ def trainModelRl(debug, modelDir, filename, nbIter, batchSize, devFile, transiti
             policy_net.train()
         i += 1
 
-        if state is None :
+        if state is None or count == countBreak:
           break
       if i >= nbExByEpoch :
         break

main.py

@@ -41,7 +41,7 @@ if __name__ == "__main__" :
   parser.add_argument("--gamma", default=0.99,
     help="Importance given to future rewards.")
   parser.add_argument("--bootstrap", default=None,
-    help="If not none, extract examples in bootstrap mode (oracle train only).")
+    help="If not none, extract examples in bootstrap mode every n epochs (oracle train only).")
   parser.add_argument("--dev", default=None,
     help="Name of the CoNLL-U file of the dev corpus.")
   parser.add_argument("--incr", "-i", default=False, action="store_true",
@@ -51,7 +51,7 @@ if __name__ == "__main__" :
   parser.add_argument("--silent", "-s", default=False, action="store_true",
     help="Don't print advancement infos.")
   parser.add_argument("--transitions", default="eager",
-    help="Transition set to use (eager | swift).")
+    help="Transition set to use (eager | swift | tagparser).")
   parser.add_argument("--ts", default="",
     help="Comma separated list of supplementary transitions. Example \"BACK 1,BACK 2\"")
   parser.add_argument("--reward", default="A",
@@ -60,6 +60,8 @@ if __name__ == "__main__" :
     help="Evolution of probability to chose action at random : (start value, decay speed, end value)")
   parser.add_argument("--probaOracle", default="0.3,2,0.0",
     help="Evolution of probability to chose action from oracle : (start value, decay speed, end value)")
+  parser.add_argument("--countBreak", default=1,
+    help="Number of unaplayable transition picked before breaking the analysis.")
   args = parser.parse_args()
 
   if args.debug :
@@ -99,7 +101,7 @@ if __name__ == "__main__" :
     json.dump(strategy, open(args.model+"/strategy.json", "w"))
     printTS(transitionSet, sys.stderr)
     probas = [list(map(float, args.probaRandom.split(','))), list(map(float, args.probaOracle.split(',')))]
-    Train.trainMode(args.debug, args.corpus, args.type, transitionSet, strategy, args.model, int(args.iter), int(args.batchSize), args.dev, args.bootstrap, args.incr, args.reward, float(args.lr), float(args.gamma), probas, args.predicted, args.silent)
+    Train.trainMode(args.debug, args.corpus, args.type, transitionSet, strategy, args.model, int(args.iter), int(args.batchSize), args.dev, args.bootstrap, args.incr, args.reward, float(args.lr), float(args.gamma), probas, int(args.countBreak), args.predicted, args.silent)
   elif args.mode == "decode" :
     transNames = json.load(open(args.model+"/transitions.json", "r"))
     transitionSet = [Transition(elem) for elem in transNames]