Added bootstrap mode for oracle training

Train.py

@@ -15,14 +15,14 @@ import Config
 from conll18_ud_eval import load_conllu, evaluate
 
 ################################################################################
-def trainMode(debug, filename, type, modelDir, nbIter, batchSize, devFile, silent=False) :
+def trainMode(debug, filename, type, modelDir, nbIter, batchSize, devFile, bootstrapInterval, silent=False) :
   transitionSet = [Transition(elem) for elem in ["RIGHT", "LEFT", "SHIFT", "REDUCE"]]
   strategy = {"RIGHT" : 1, "SHIFT" : 1, "LEFT" : 0, "REDUCE" : 0}
 
   sentences = Config.readConllu(filename)
 
   if type == "oracle" :
-    trainModelOracle(debug, modelDir, filename, nbIter, batchSize, devFile, transitionSet, strategy, sentences, silent)
+    trainModelOracle(debug, modelDir, filename, nbIter, batchSize, devFile, transitionSet, strategy, sentences, bootstrapInterval, silent)
     return
 
   if type == "rl":
@@ -34,25 +34,36 @@ def trainMode(debug, filename, type, modelDir, nbIter, batchSize, devFile, silen
 ################################################################################
 
 ################################################################################
-def extractExamples(ts, strat, config, dicts, debug=False) :
+def extractExamples(debug, ts, strat, config, dicts, network=None) :
   examples = []
-
+  with torch.no_grad() :
     EOS = Transition("EOS")
     config.moveWordIndex(0)
     moved = True
     while moved :
       missingLinks = getMissingLinks(config)
-    candidates = sorted([[trans.getOracleScore(config, missingLinks), trans.name] for trans in ts if trans.appliable(config)])
+      candidates = sorted([[trans.getOracleScore(config, missingLinks), trans] for trans in ts if trans.appliable(config)])
       if len(candidates) == 0 :
         break
-    candidate = candidates[0][1]
-    candidateIndex = [trans.name for trans in ts].index(candidate)
+      best = min([cand[0] for cand in candidates])
+      candidateOracle = random.sample([cand for cand in candidates if cand[0] == best], 1)[0][1]
       features = Features.extractFeatures(dicts, config)
-    example = torch.cat([torch.LongTensor([candidateIndex]), features])
-    examples.append(example)
+      candidate = candidateOracle.name
       if debug :
         config.printForDebug(sys.stderr)
-      print(str(candidates)+"\n"+("-"*80)+"\n", file=sys.stderr)
+        print(str([[c[0],c[1].name] for c in candidates])+"\n"+("-"*80)+"\n", file=sys.stderr)
+      if network is not None :
+        output = network(features.unsqueeze(0).to(getDevice()))
+        scores = sorted([[float(output[0][index]), ts[index].appliable(config), ts[index].name] for index in range(len(ts))])[::-1]
+        candidate = [[cand[0],cand[2]] for cand in scores if cand[1]][0][1]
+        if debug :
+          print(candidate.name, file=sys.stderr)
+        
+      goldIndex = [trans.name for trans in ts].index(candidateOracle.name)
+      candidateIndex = [trans.name for trans in ts].index(candidate)
+      example = torch.cat([torch.LongTensor([goldIndex]), features])
+      examples.append(example)
+
       moved = applyTransition(ts, strat, config, candidate)
   
     EOS.apply(config)
@@ -82,14 +93,15 @@ def evalModelAndSave(debug, model, dicts, modelDir, devFile, bestLoss, totalLoss
 ################################################################################
 
 ################################################################################
-def trainModelOracle(debug, modelDir, filename, nbEpochs, batchSize, devFile, transitionSet, strategy, sentences, silent=False) :
-  examples = []
+def trainModelOracle(debug, modelDir, filename, nbEpochs, batchSize, devFile, transitionSet, strategy, sentencesOriginal, bootstrapInterval, silent=False) :
   dicts = Dicts()
   dicts.readConllu(filename, ["FORM", "UPOS"])
   dicts.save(modelDir+"/dicts.json")
+  examples = []
+  sentences = copy.deepcopy(sentencesOriginal)
   print("%s : Starting to extract examples..."%(timeStamp()), file=sys.stderr)
   for config in sentences :
-    examples += extractExamples(transitionSet, strategy, config, dicts, debug)
+    examples += extractExamples(debug, transitionSet, strategy, config, dicts)
   print("%s : Extracted %s examples"%(timeStamp(), prettyInt(len(examples), 3)), file=sys.stderr)
   examples = torch.stack(examples)
 
@@ -100,6 +112,15 @@ def trainModelOracle(debug, modelDir, filename, nbEpochs, batchSize, devFile, tr
   bestLoss = None
   bestScore = None
   for epoch in range(1,nbEpochs+1) :
+    if bootstrapInterval is not None and epoch > 1 and (epoch-1) % bootstrapInterval == 0 :
+      examples = []
+      sentences = copy.deepcopy(sentencesOriginal)
+      print("%s : Starting to extract dynamic examples..."%(timeStamp()), file=sys.stderr)
+      for config in sentences :
+        examples += extractExamples(debug, transitionSet, strategy, config, dicts, network)
+      print("%s : Extracted %s examples"%(timeStamp(), prettyInt(len(examples), 3)), file=sys.stderr)
+      examples = torch.stack(examples)
+
     network.train()
     examples = examples.index_select(0, torch.randperm(examples.size(0)))
     totalLoss = 0.0

main.py

@@ -27,6 +27,8 @@ if __name__ == "__main__" :
     help="Size of each batch.")
   parser.add_argument("--seed", default=100,
     help="Random seed.")
+  parser.add_argument("--bootstrap", default=None,
+    help="If not none, extract examples in bootstrap mode (oracle train only).")
   parser.add_argument("--dev", default=None,
     help="Name of the CoNLL-U file of the dev corpus.")
   parser.add_argument("--debug", "-d", default=False, action="store_true",
@@ -43,7 +45,7 @@ if __name__ == "__main__" :
   torch.manual_seed(args.seed)
 
   if args.mode == "train" :
-    Train.trainMode(args.debug, args.corpus, args.type, args.model, int(args.iter), int(args.batchSize), args.dev, args.silent)
+    Train.trainMode(args.debug, args.corpus, args.type, args.model, int(args.iter), int(args.batchSize), args.dev, args.bootstrap, args.silent)
   elif args.mode == "decode" :
     Decode.decodeMode(args.debug, args.corpus, args.type, args.model)
   else :