Ported code of BiLSTM to Keras 2.0.5

2017-07_Seminar/Session 1 - SENNA/code for NER/NER_solution.py → 2017-07_Seminar/Session 1 - SENNA/code for NER/NER.py

@@ -70,16 +70,16 @@
 
 
 # Create the train and predict_labels function
-n_in = train_tokens.shape[1]
-n_out = len(label2Idx)
+n_in = train_tokens.shape[1] # Length of the input, i.e. 7 for a window size of 7
+n_out = len(label2Idx) # Number of output labels
 
 words_input = Input(shape=(n_in,), dtype='int32', name='words_input')
-words = Embedding(input_dim=wordEmbeddings.shape[0], output_dim=wordEmbeddings.shape[1], input_length=n_in, weights=[wordEmbeddings], trainable=False)(words_input)
+words = Embedding(input_dim=wordEmbeddings.shape[0], output_dim=wordEmbeddings.shape[1], weights=[wordEmbeddings], trainable=False)(words_input)
 words = Flatten()(words)
 
 
 casing_input = Input(shape=(n_in,), dtype='int32', name='casing_input')
-casing = Embedding(input_dim=caseEmbeddings.shape[0], output_dim=caseEmbeddings.shape[1], input_length=n_in, weights=[caseEmbeddings], trainable=False)(casing_input)
+casing = Embedding(input_dim=caseEmbeddings.shape[0], output_dim=caseEmbeddings.shape[1], weights=[caseEmbeddings], trainable=False)(casing_input)
 casing = Flatten()(casing)
 
 output = concatenate([words, casing])

2017-07_Seminar/Session 1 - SENNA/code for POS/POS_solution.py → 2017-07_Seminar/Session 1 - SENNA/code for POS/POS.py

@@ -66,16 +66,16 @@
 
 
 # Create the train and predict_labels function
-n_in = train_tokens.shape[1]
-n_out = len(label2Idx)
+n_in = train_tokens.shape[1] # Length of the input, i.e. 7 for a window size of 7
+n_out = len(label2Idx) # Number of output labels
 
 words_input = Input(shape=(n_in,), dtype='int32', name='words_input')
-words = Embedding(input_dim=wordEmbeddings.shape[0], output_dim=wordEmbeddings.shape[1], input_length=n_in, weights=[wordEmbeddings], trainable=False)(words_input)
+words = Embedding(input_dim=wordEmbeddings.shape[0], output_dim=wordEmbeddings.shape[1], weights=[wordEmbeddings], trainable=False)(words_input)
 words = Flatten()(words)
 
 
 casing_input = Input(shape=(n_in,), dtype='int32', name='casing_input')
-casing = Embedding(input_dim=caseEmbeddings.shape[0], output_dim=caseEmbeddings.shape[1], input_length=n_in, weights=[caseEmbeddings], trainable=False)(casing_input)
+casing = Embedding(input_dim=caseEmbeddings.shape[0], output_dim=caseEmbeddings.shape[1], weights=[caseEmbeddings], trainable=False)(casing_input)
 casing = Flatten()(casing)
 
 output = concatenate([words, casing])
@@ -115,11 +115,12 @@ def predict_classes(prediction):
     print("\n------------- Epoch %d ------------" % (epoch+1))
     model.fit([train_tokens, train_case], train_y, epochs=1, batch_size=minibatch_size, verbose=True, shuffle=True)   
 
-
+    #Predict labels for development set
     dev_pred = predict_classes(model.predict([dev_tokens, dev_case]))
     dev_acc = np.sum(dev_pred == dev_y) / float(len(dev_y))
     print("Dev-Accuracy: %.2f" % (dev_acc*100))
 
+    #Predict labels for test set
     test_pred = predict_classes(model.predict([test_tokens, test_case]))
     test_acc = np.sum(test_pred == test_y) / float(len(test_y))
     print("Test-Accuracy: %.2f" % (test_acc*100))

2017-07_Seminar/Session 1 - SENNA/code for POS/preprocess.py

@@ -187,7 +187,7 @@ def getCasing(word, caseLookup):
     basename = os.path.basename(embeddingsPath)
     if basename == 'wiki_extvec.gz':
 	       print("Start downloading word embeddings for English using wget ...")
-	       #os.system("wget https://www.cs.york.ac.uk/nlp/extvec/"+basename+" -P embeddings/")
+	       #os.system("wget https://www.cs.york.ac.uk/nlp/extvec/"+basename+" -P embeddings/") #Original path from York University
 	       os.system("wget https://public.ukp.informatik.tu-darmstadt.de/reimers/2017_english_embeddings/"+basename+" -P embeddings/")
     else:
         print(embeddingsPath, "does not exist. Please provide pre-trained embeddings")

2017-07_Seminar/Session 2 - Sentence CNN/code/cnn_solution.py → 2017-07_Seminar/Session 2 - Sentence CNN/code/cnn.py

@@ -128,8 +128,7 @@ def wordIdxLookup(word, word_idx_map):
 
 #Our word embedding layer
 wordsEmbeddingLayer = Embedding(word_embeddings.shape[0],
-                    word_embeddings.shape[1],
-                    input_length=max_sentence_len,                   
+                    word_embeddings.shape[1],                                     
                     weights=[word_embeddings],
                     trainable=False)
 
@@ -152,7 +151,7 @@ def wordIdxLookup(word, word_idx_map):
 
 
 
-# We add a vanilla hidden layer:
+# We add a vanilla hidden layer together with dropout layers:
 output = Dropout(0.5)(output)
 output = Dense(hidden_dims, activation='tanh', kernel_regularizer=keras.regularizers.l2(0.01))(output)
 output = Dropout(0.25)(output)
@@ -162,16 +161,15 @@ def wordIdxLookup(word, word_idx_map):
 output = Dense(1, activation='sigmoid',  kernel_regularizer=keras.regularizers.l2(0.01))(output)
 
 model = Model(inputs=[words_input], outputs=[output])
-model.compile(loss='binary_crossentropy',
-              optimizer='adam',
-              metrics=['accuracy'])
+model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
 
 model.summary()
 
 for epoch in range(nb_epoch):
     print("\n------------- Epoch %d ------------" % (epoch+1))
     model.fit(X_train, y_train, batch_size=batch_size, epochs=1)
 
+    #Use Keras to compute the loss and the accuracy
     dev_loss, dev_accuracy = model.evaluate(X_dev, y_dev, verbose=False)
     test_loss, test_accuracy = model.evaluate(X_test, y_test, verbose=False)