debug

expe/vazy.sh

@@ -13,7 +13,6 @@ dev_cff="./out/dev_${lang}.cff"
 dev_word_limit="5000"
 
 test_conll="../data/test_${lang}.conllu"
-#test_conll="../data/test_${lang}_5sent.conllu"
 
 test_mcf="./out/test_${lang}_pgle.mcf"
 test_mcf_hyp="./out/test_${lang}_hyp.mcf"

src/tbp_decode_pytorch.py

@@ -39,8 +39,6 @@ train_fr_file = sys.argv[7]
 dev_fr_file = sys.argv[8]
 
 ##########################################################################
-#dev_fr_file = '/home/taniabladier/Programming/AMU/tbp/expe/out/dev_fr1.cff'
-#train_fr_file = '/home/taniabladier/Programming/AMU/tbp/expe/out/train_fr1.cff'
 
 n_classes, maxlen, n_symbols, symbol_to_idx, idx_to_symbol, class_to_idx, idx_to_class, _ = make_pytorch_dicts(dev_fr_file, train_fr_file)
 ##########################################################################
@@ -102,16 +100,16 @@ class SimpleLSTM(nn.Module):
 
 
 
-input_size  = n_symbols
+input_size  = 133 #n_symbols
 hidden_size = 128
-output_size = n_classes
+output_size = 75 #n_classes
 
+
+print('input output', input_size, output_size)
 model       = SimpleLSTM(input_size, hidden_size, output_size)
 criterion   = torch.nn.CrossEntropyLoss()
 optimizer   = torch.optim.RMSprop(model.parameters(), lr=0.001)
 
-#model_file = '/home/taniabladier/Programming/AMU/tbp/expe/out/fr.pytorch'
-#mcf_file = '/home/taniabladier/Programming/AMU/tbp/expe/out/test_fr_pgle.mcf'
 
 checkpoint = torch.load(model_file, map_location=torch.device('cpu'), weights_only=False)
 model.load_state_dict(checkpoint['model_state_dict'])
@@ -130,8 +128,8 @@ model.eval();
 
 #model = load_model(model_file)
 
-inputSize = featModel.getInputSize()
-outputSize = moves.getNb()
+#inputSize = featModel.getInputSize()
+#outputSize = moves.getNb()
 
 c = Config(mcf_file, mcd, dicos)
 
@@ -154,7 +152,7 @@ while c.getBuffer().readNextSentence()  and numWords < wordsLimit :
 
         ###############
         inputVector = ' '.join(str(x) for x in inputVector)
-        inputVector = encode_x_batch([inputVector], symbol_to_idx, n_symbols)
+        inputVector = encode_x_batch([inputVector], symbol_to_idx, 133) #n_symbols)
         inputVector = torch.from_numpy(inputVector).float().to(device)      
        
 
@@ -209,10 +207,6 @@ while c.getBuffer().readNextSentence()  and numWords < wordsLimit :
     for i in range(1, c.getBuffer().getLength()):
         w = c.getBuffer().getWord(i)
         w.affiche(mcd)
-        #print('')
-        #print('5555\t', w.getFeat("GOV"), end='\t')
-        #print('\n5566\t', w.getFeat("LABEL"))
-        #print('\n5566\t', w.getFeat("POS"))
 
     numSent += 1
 #    if numSent % 10 == 0:

src/tbp_train_pytorch.py

@@ -5,7 +5,7 @@ import torch.nn as nn
 import torch.nn.functional as F
 from pytorch_utils import *
 from plot_lib import *
-
+import os
 
 """## 1. Reading Data Files"""
 
@@ -67,17 +67,9 @@ n_classes, maxlen, n_symbols, symbol_to_idx, idx_to_symbol, \
 train_items_list, train_labels_list, train_inputSize, train_outputSize  = readFile_cff(cffTrainFileName)
 dev_items_list, dev_labels_list, dev_inputSize, dev_outputSize  = readFile_cff(cffDevFileName)
 
-#print(len(train_items_list))
-#print(train_items_list[:3])
-#print(len(dev_items_list))
 
 train_data_gen = preprocess_data(train_items_list[:800000], train_labels_list[:800000], batch_size, symbol_to_idx, class_to_idx, train_inputSize, train_outputSize)#, n_symbols, n_classes)
 dev_data_gen  = preprocess_data(dev_items_list[:200000], dev_labels_list[:200000], batch_size, symbol_to_idx, class_to_idx, train_inputSize, train_outputSize)#, n_symbols, n_classes)
-#train_data_gen = preprocess_data(train_items_list[:800], train_labels_list[:800])
-#dev_data_gen  = preprocess_data(dev_items_list[:200], dev_labels_list[:200])
-
-#print(len(train_items_list))
-
 
 
 """## 2. Defining the Model"""
@@ -87,9 +79,9 @@ dev_data_gen  = preprocess_data(dev_items_list[:200000], dev_labels_list[:200000
 torch.manual_seed(1)
 
 # Setup the RNN and training settings
-input_size  = train_inputSize #n_symbols
+input_size  = 133 #train_inputSize #n_symbols
 hidden_size = 128
-output_size = train_outputSize #n_classes
+output_size = 75 #train_outputSize #n_classes
 
 class SimpleMLP(nn.Module):
     def __init__(self, input_size, output_size):
@@ -157,44 +149,6 @@ class SimpleLSTM(nn.Module):
             c = torch.cat(c_list)
         return h, c
     
-class BiLSTM(nn.Module):
-    
-    def __init__(self, input_size, hidden_size, output_size):
-        super().__init__()
-        #self.hidden_size = 64
-        
-        self.lstm = nn.LSTM(input_size, hidden_size, bidirectional=True, batch_first=True)
-        self.linear = nn.Linear(hidden_size*4 , output_size)
-        self.relu = nn.ReLU()
-        self.dropout = nn.Dropout(0.5)
-        self.out = nn.Linear(output_size, 1)
-
-
-    def forward(self, x):
-        
-        h_lstm = self.lstm(x)[0]
-        avg_pool = torch.mean(h_lstm, 1)
-        max_pool, _ = torch.max(h_lstm, 1)
-        #print("avg_pool", avg_pool.size())
-        #print("max_pool", max_pool.size())
-        conc = torch.cat(( avg_pool, max_pool), 1)
-        conc = self.relu(self.linear(conc))
-        conc = self.dropout(conc)
-        out = self.out(conc)
-        return out
-    
-    def get_states_across_time(self, x):
-        h_c = None
-        h_list, c_list = list(), list()
-        with torch.no_grad():
-            for t in range(out.size(1)):
-                h_c = self.lstm(x[:, [t], :], h_c)[1]
-                h_list.append(h_c[0])
-                c_list.append(h_c[1])
-            h = torch.cat(h_list)
-            c = torch.cat(c_list)
-        return h, c
-        
 
 """## 3. Defining the Training Loop"""
 
@@ -359,7 +313,7 @@ def train_and_test(model, train_data_gen, test_data_gen, criterion, optimizer, m
         ax.set_xlabel('epoch', fontsize=12)
         ax.set_ylabel(metric, fontsize=12)
         ax.legend(['Train', 'Test'], loc='best')
-        plt.savefig('./expe/out/loss_accuracy.png')
+        plt.savefig(os.path.abspath('..') + '/expe/out/loss_accuracy.png')
     #plt.show()
 
     return model
@@ -425,28 +379,10 @@ max_epochs  = 30
 # Train the model
 model = train_and_test(model, train_data_gen, dev_data_gen, criterion, optimizer, max_epochs)
 
-for parameter_group in list(model.parameters()):
-    print(parameter_group.size())
-
-
-""" 6b. BiLSTM
+#for parameter_group in list(model.parameters()):
+#    print(parameter_group.size())
 
-"""
-
-"""
-model       = BiLSTM(input_size, hidden_size, output_size)
-criterion   = torch.nn.CrossEntropyLoss()
-optimizer   = torch.optim.RMSprop(model.parameters(), lr=0.001)
 
-
-max_epochs  = 30
-
-# Train the model
-model = train_and_test(model, train_data_gen, dev_data_gen, criterion, optimizer, max_epochs)
-
-for parameter_group in list(model.parameters()):
-    print(parameter_group.size())
-"""
 """## 7. Model Evaluation"""
 
 import collections
@@ -527,6 +463,7 @@ def evaluate_model(model, seed=9001, verbose=False):
 
 evaluate_model(model)
 
+
 """ Visualize Model """
 
 # Get hidden (H) and cell (C) batch state given a batch input (X)