some corr

.gitignore

@@ -3,3 +3,4 @@ datasets
 *.npy
 *.npz
 *.pyc
+embedding_weights/*

README.md

@@ -4,7 +4,7 @@ Pytorch implementation of the paper http://nyc.lti.cs.cmu.edu/yiming/Publication
 ## Dependencies
 
 * NLTK (stopwords)
-* Pytorch
+* Pytorch >= 0.3.1
 * Gensim
 * Matplotlib
 

code/cnn_test.py

@@ -44,7 +44,7 @@ def test_class(x_te, y_te, params, model=None, x_tr=None, y_tr=None, embedding_w
         prec = precision_k(y_tr, Y, 5)
         print('Test Loss; Precision Scores [1->5] {} Cross Entropy {};'.format(prec, loss))
 
-    y_te = y_te[:,:-1]
+    #y_te = y_te[:,:-1]
     x_te, _ = load_batch_cnn(x_te, y_te, params, batch=False)
     Y2 = np.zeros(y_te.shape)
     rem = x_te.shape[0]%params.mb_size
@@ -67,4 +67,4 @@ def test_class(x_te, y_te, params, model=None, x_tr=None, y_tr=None, embedding_w
         Y_probabs2 = sparse.csr_matrix(Y2)
         sio.savemat('score_matrix.mat' , {'score_matrix': Y_probabs2})
 
-    return prec[0], loss
+    return prec[0], loss

code/cnn_train.py

@@ -76,8 +76,8 @@ def train(x_tr, y_tr, x_te, y_te, embedding_weights, params):
 					win = viz.line(X=np.arange(iteration, iteration + .1), Y=np.arange(0, .1))
 			iteration +=1
 
-			if(epoch==0):
-				break
+			#if(epoch==0):
+			#	break
 
 		if(totalLoss<bestTotalLoss):
 

utils/w2v.py

@@ -16,7 +16,7 @@ def train_word2vec(sentence_matrix, vocabulary_inv,
     min_word_count  # Minimum word count
     context         # Context window size
     """
-    model_dir = 'word2vec_models'
+    model_dir = '../embedding_weights'
     model_name = "{:d}features_{:d}minwords_{:d}context".format(num_features, min_word_count, context)
     model_name = join(model_dir, model_name)
     if exists(model_name):
@@ -62,7 +62,7 @@ def load_word2vec(params):
     num_features    # Word vector dimensionality
     """
 
-    model_dir = 'word2vec_models'
+    model_dir = '../embedding_weights'
 
     if params.model_type == 'GoogleNews':
         model_name = join(model_dir, 'GoogleNews-vectors-negative300.bin.gz')
@@ -73,7 +73,8 @@ def load_word2vec(params):
 
     elif params.model_type == 'glove':
         model_name = join(model_dir, 'glove.6B.%dd.txt' % (params.num_features))
-        assert(exists(model_name))
+        print(model_name)
+	assert(exists(model_name))
         print('Loading existing Word2Vec model (Glove.6B.%dd)' % (params.num_features))
 
         # dictionary, where key is word, value is word vectors