updated for italian version

grut.py

@@ -58,7 +58,7 @@ def defineTrainArguments(n_fold, use_cuda, epochs, targetType, modelName, modelV
 
     # Optional argument
     parser.add_argument('-mn','--model_name', type=str,
-                        help='model name. Default "' + str(modelName) + '". Define only in train mode.') 
+                        help='model name. Choices: ["bart", "mbart", "t5", "mt5"], Default "' + str(modelName) + '". Define only in train mode.') 
 
     # Optional argument
     parser.add_argument('-mv','--model_variant', type=str,

huricParser.py

@@ -300,7 +300,7 @@ def parseMap(self, entities, lexicalGroundings, lexicalizedMap: bool):
 
 
     def computeLexicalGroundingsANDparseMap(self, entities, sentence, map_type, entityRetrievalType, lexicalReferences = "all", thresholdW2V = 0.5, thresholdLDIST = 0.8):
-        print("computeLexicalGroundingsANDparseMap")
+        # print("computeLexicalGroundingsANDparseMap")
 
         entities_list = []
         atoms = {}

requirements.txt

@@ -3,7 +3,7 @@ numpy==1.22.2
 pandas==1.4.1
 scikit_learn==1.1.3
 seqeval==1.2.2
-simpletransformers==0.63.4
+simpletransformers==0.63.9
 spacy==3.3.1
 torch==1.10.2
 transformers==4.16.2

train.sh

@@ -1,9 +1,9 @@
 #!/bin/sh
 # BART_base
-    nohup python -u main.py train -mn bart -mv base -t SRL -bs 32 -uc True -ep 50 -nf 10 -tt SRL > logs/testing_verbose_with_prints_bart_base.out
+    nohup python -u grut.py train -mn bart -mv base -t SRL -bs 32 -uc True -ep 50 -nf 10 -tt SRL > logs/testing_verbose_with_prints_bart_base.out
 
 # GRUT
-    nohup python -u main.py train -mn bart -mv base -t SRL -bs 32 -uc True -ep 50 -nf 10 -tt SRL -gr half -map lmd > logs/testing_verbose_with_prints_grut.out
+    nohup python -u grut.py train -mn bart -mv base -t SRL -bs 32 -uc True -ep 50 -nf 10 -tt SRL -gr half -map lmd > logs/testing_verbose_with_prints_grut.out
 
 # GRUT_LU
-    nohup python -u main.py train -mn bart -mv base -t SRL -bs 32 -uc True -ep 50 -nf 10 -tt SRL -gr half -map lmd -alut True > logs/testing_verbose_with_prints_grut_lu.out
+    nohup python -u grut.py train -mn bart -mv base -t SRL -bs 32 -uc True -ep 50 -nf 10 -tt SRL -gr half -map lmd -alut True > logs/testing_verbose_with_prints_grut_lu.out

trainer.py

@@ -52,15 +52,18 @@ def __init__(self, lan: Language, model = 'bart', model_variant = 'base', task =
             self.model_args.gradient_accumulation_steps = 2
             self.model_args.early_stopping_patience = patience
             self.model_args.early_stopping_delta = 1e-4 #0.005
-        elif model == "bart":
+        elif model == "bart" or model == "mbart":
             self.model_args = Seq2SeqArgs()
             self.model_args.fp16 = False
             self.model_args.optimizer = "AdamW" #"Adafactor"
             self.model_args.scheduler = "linear_schedule_with_warmup" #"constant_schedule" #"constant_schedule_with_warmup" #
             self.model_args.learning_rate = learning_rate
             self.model_args.eval_batch_size = batch_size
             self.model_args.train_batch_size = batch_size
-            self.model_name = 'facebook/bart-' + model_variant
+            if model == "bart":
+                self.model_name = 'facebook/bart-' + model_variant
+            else:
+                self.model_name = 'facebook/mbart-large-cc25'
             self.model_args.gradient_accumulation_steps = 2
             self.model_args.early_stopping_patience = patience
             self.model_args.early_stopping_delta = 1e-4 #0.001
@@ -111,7 +114,7 @@ def train_saving_all_folds_models(self, num_folds, quick_train = False, addMap =
         else:
             lutype = ""
 
-        kfold_folder_name = self.model + '_' + self.lan.value + "_" + map_type + "_" + grounding + "grounding_" + entityRetrievalType + "_" + lexicalReferences + "LexicalReferences" + lutype
+        kfold_folder_name = self.lan.value + "/" + self.model + '_' + "_" + map_type + "_" + grounding + "grounding_" + entityRetrievalType + "_" + lexicalReferences + "LexicalReferences" + lutype
         kfold_folder_name += "_" + date_time
         kfold_base_dir = model_folder_name + '/' + kfold_folder_name + '/'
         hp = HuricParser(self.lan)
@@ -125,7 +128,6 @@ def train_saving_all_folds_models(self, num_folds, quick_train = False, addMap =
             hp.writeHuricSentences("./data/huric/", "./data/huric_sentences_" + self.lan.value + ".csv")
 
         df = pd.read_csv("./data/huric_sentences_" + self.lan.value + ".csv")
-
 
         # if quick_train is True, take 100 examples and quick train
         if quick_train:
@@ -258,13 +260,16 @@ def train_saving_all_folds_models(self, num_folds, quick_train = False, addMap =
                 eval_df['prefix'] = [self.task] * len(eval_df['input_text'].tolist())
                 train_df['prefix'] = [self.task] * len(train_df['input_text'].tolist())
 
-            elif self.model in ["bart"]:
+            elif self.model in ["bart", "mbart"]:
                 model = Seq2SeqModel(
-                    encoder_decoder_type="bart",
+                    encoder_decoder_type=self.model,
                     encoder_decoder_name=self.model_name,
                     args=self.model_args,
                     use_cuda=cuda_available
                 )
+
+            print(allData.head())
+            quit()
 
             print("_______________STARTING TO TRAIN_______________")
             model_train_out = model.train_model(train_df, eval_data=eval_df)