minor fix

cnn.py

@@ -62,15 +62,15 @@
 vocab, reverse_vocab = {}, {}
 freq = defaultdict(int)
 tweets = {}
-
+np.random.seed(42)
 
 def get_embedding(word):
     #return
     try:
         return word2vec_model[word]
     except Exception, e:
         print 'Encoding not found: %s' %(word)
-        return np.zeros(EMBEDDING_DIM) 
+        return np.zeros(EMBEDDING_DIM)
 
 def get_embedding_weights():
     embedding = np.zeros((len(vocab) + 1, EMBEDDING_DIM))
@@ -154,7 +154,7 @@ def gen_sequence():
         y.append(y_map[tweet['label']])
     return X, y
 
-    
+
 def Tokenize(tweet):
     #return MyTokenizer.tokenize(tweet)
     #pdb.set_trace()
@@ -182,8 +182,8 @@ def cnn_model(sequence_length, embedding_dim):
 
     # Training parameters
     # Word2Vec parameters, see train_word2vec
-    #min_word_count = 1  # Minimum word count                        
-    #context = 10        # Context window size    
+    #min_word_count = 1  # Minimum word count
+    #context = 10        # Context window size
 
     graph_in = Input(shape=(sequence_length, embedding_dim))
     convs = []
@@ -196,7 +196,7 @@ def cnn_model(sequence_length, embedding_dim):
         pool = GlobalMaxPooling1D()(conv)
         #flatten = Flatten()(pool)
         convs.append(pool)
-        
+
     if len(filter_sizes)>1:
         out = Merge(mode='concat')(convs)
     else:
@@ -277,7 +277,7 @@ def train_CNN(X, y, inp_dim, model, weights, epochs=10, batch_size=128):
     tweets = Tweets
     gen_vocab()
     #filter_vocab(20000)
-    X, y = gen_sequence()    
+    X, y = gen_sequence()
     #Y = y.reshape((len(y), 1))
     MAX_SEQUENCE_LENGTH = max(map(lambda x:len(x), X))
     print "max seq length is %d"%(MAX_SEQUENCE_LENGTH)
@@ -287,7 +287,7 @@ def train_CNN(X, y, inp_dim, model, weights, epochs=10, batch_size=128):
     W = get_embedding_weights()
     model = cnn_model(data.shape[1], EMBEDDING_DIM)
     train_CNN(data, y, EMBEDDING_DIM, model, W)
-    
+
     pdb.set_trace()
 
 

fast_text.py

@@ -38,7 +38,7 @@
 print('Found %s texts. (samples)' % len(texts))
 
 EMBEDDING_DIM = int(sys.argv[1])
-
+np.random.seed(42)
 # Load the orginal glove file
 # SHASHANK files
 #GLOVE_MODEL_FILE="/home/shashank/DL_NLP/glove-twitter" + str(EMBEDDING_DIM) + "-w2v"
@@ -66,7 +66,7 @@ def get_embedding(word):
         return word2vec_model[word]
     except Exception, e:
         print 'Encoding not found: %s' %(word)
-        return np.zeros(EMBEDDING_DIM) 
+        return np.zeros(EMBEDDING_DIM)
 
 def get_embedding_weights():
     embedding = np.zeros((len(vocab) + 1, EMBEDDING_DIM))
@@ -150,7 +150,7 @@ def gen_sequence():
         y.append(y_map[tweet['label']])
     return X, y
 
-    
+
 def Tokenize(tweet):
     #return MyTokenizer.tokenize(tweet)
     #pdb.set_trace()
@@ -217,7 +217,7 @@ def train_fasttext(X, y, model, inp_dim,embedding_weights, epochs=10, batch_size
         r1 += recall_score(y_test, y_pred, average='micro')
         f1 += f1_score(y_test, y_pred, average='weighted')
         f11 += f1_score(y_test, y_pred, average='micro')
-    
+
     print "macro results are"
     print "average precision is %f" %(p/10)
     print "average recall is %f" %(r/10)
@@ -235,7 +235,7 @@ def check_semantic_sim(embedding_table, word):
     sim_word_idx = get_similar_words(embedding_table, embedding_table[vocab[word]], 25)
     sim_words = map(lambda x:reverse_vocab[x[1]], sim_word_idx)
     print sim_words
-        
+
 def tryWord(embedding_table):
     while True:
         print "enter word"
@@ -253,15 +253,15 @@ def tryWord(embedding_table):
     Tweets = select_tweets()
     tweets = Tweets
     gen_vocab()
-    X, y = gen_sequence()    
+    X, y = gen_sequence()
     MAX_SEQUENCE_LENGTH = max(map(lambda x:len(x), X))
     print "max seq length is %d"%(MAX_SEQUENCE_LENGTH)
     data = pad_sequences(X, maxlen=MAX_SEQUENCE_LENGTH)
     y = np.array(y)
     W = get_embedding_weights()
     data, y = sklearn.utils.shuffle(data, y)
     model = fast_text_model(data.shape[1])
-    _ = train_fasttext(data, y, model, EMBEDDING_DIM, W)    
+    _ = train_fasttext(data, y, model, EMBEDDING_DIM, W)
     table = model.layers[0].get_weights()[0]
     #check_semantic_sim(table)
     tryWord(table)

lstm.py

@@ -50,7 +50,7 @@
 MAX_NB_WORDS = None
 VALIDATION_SPLIT = 0.2
 word2vec_model = gensim.models.Word2Vec.load_word2vec_format(GLOVE_MODEL_FILE)
-
+np.random.seed(42)
 
 # vocab generation
 MyTokenizer = tokenize.casual.TweetTokenizer(strip_handles=True, reduce_len=True)
@@ -65,7 +65,7 @@ def get_embedding(word):
         return word2vec_model[word]
     except Exception, e:
         print 'Encoding not found: %s' %(word)
-        return np.zeros(EMBEDDING_DIM) 
+        return np.zeros(EMBEDDING_DIM)
 
 def get_embedding_weights():
     embedding = np.zeros((len(vocab) + 1, EMBEDDING_DIM))
@@ -149,7 +149,7 @@ def gen_sequence():
         y.append(y_map[tweet['label']])
     return X, y
 
-    
+
 def Tokenize(tweet):
     #return MyTokenizer.tokenize(tweet)
     #pdb.set_trace()
@@ -204,7 +204,7 @@ def train_LSTM(X, y, model, inp_dim, weights, epochs=10, batch_size=512):
                 print x.shape, y.shape
                 loss, acc = model.train_on_batch(x, y_temp)#, class_weight=class_weights)
                 print loss, acc
-        
+
         y_pred = model.predict_on_batch(X_test)
         y_pred = np.argmax(y_pred, axis=1)
         print classification_report(y_test, y_pred)
@@ -216,7 +216,7 @@ def train_LSTM(X, y, model, inp_dim, weights, epochs=10, batch_size=512):
         r1 += recall_score(y_test, y_pred, average='micro')
         f1 += f1_score(y_test, y_pred, average='weighted')
         f11 += f1_score(y_test, y_pred, average='micro')
-	
+
 
     print "macro results are"
     print "average precision is %f" %(p/10)
@@ -234,7 +234,7 @@ def train_LSTM(X, y, model, inp_dim, weights, epochs=10, batch_size=512):
     tweets = Tweets
     gen_vocab()
     #filter_vocab(20000)
-    X, y = gen_sequence()    
+    X, y = gen_sequence()
     #Y = y.reshape((len(y), 1))
     MAX_SEQUENCE_LENGTH = max(map(lambda x:len(x), X))
     print "max seq length is %d"%(MAX_SEQUENCE_LENGTH)
@@ -245,5 +245,5 @@ def train_LSTM(X, y, model, inp_dim, weights, epochs=10, batch_size=512):
     model = lstm_model(data.shape[1], EMBEDDING_DIM)
     #model = lstm_model(data.shape[1], 25, get_embedding_weights())
     train_LSTM(data, y, model, EMBEDDING_DIM, W)
-    
+
     pdb.set_trace()