adding debugging arguments

sm-model/README.md

@@ -20,7 +20,7 @@ $ make
 
 ``2.`` Get the Overlapping features for Q and A:
 ```
-$ python overlap_features.py TrecQA
+$ python overlap_features.py ../../data/TrecQA
 ```
 
 ``3.`` To run the S&M model on TrecQA, please follow the same parameter setting:

sm-model/main.py

@@ -67,14 +67,28 @@ if __name__ == "__main__":
     ap.add_argument('word_vectors_file', help='NOTE: a cache will be created for faster loading for word vectors')
     ap.add_argument('dataset_folder', help='directory containing train, dev, test sets')
     ap.add_argument('model_fname', help='model will be saved in args.dataset_folder/<model_fname>')
+    ap.add_argument('--classes', type=int, default=2)
+
+    # system arguments
+    # TODO: add arguments for CUDA
+    ap.add_argument('--num_threads', help="the number of simultaneous processes to run", type=int, default=4)
+
+    # training arguments
     ap.add_argument('--batch_size', type=int, default=1)
-    ap.add_argument('--filter_width', type=int, default=5)
-    ap.add_argument('--epochs', type=int, default=25)
+    ap.add_argument('--filter_width', type=int, default=5)    
     ap.add_argument('--eta', help='Initial learning rate', default=0.01, type=float)
     ap.add_argument('--mom', help='SGD Momentum', default=0.9, type=float)    
-    ap.add_argument('--classes', type=int, default=2)
+    
+    # epoch related arguments
+    ap.add_argument('--epochs', type=int, default=25)
     ap.add_argument('--patience', type=int, default=5, help="if there is no appreciable change in model after <patience> epochs, then stop")
     
+    # debugging arguments
+    ap.add_argument('--debugSingleBatch', action="store_true", help="will stop program after training 1 input batch")
+    ap.add_argument('--no_ext_feats', action="store_true", help="will not include external features in the model")
+    ap.add_argument('--num_conv_filters', help="the number of convolution channels (lesser is faster)", default=100, type=int)
+    
+
     args = ap.parse_args()
 
     torch.manual_seed(1234)
@@ -87,18 +101,20 @@ if __name__ == "__main__":
     vocab_size, vec_dim = utils.load_embedding_dimensions(cache_file)
     
     # instantiate model
-    net = QAModel(vec_dim, args.filter_width) #filter width is 5
+    net = QAModel(vec_dim, args.filter_width, args.num_conv_filters, args.no_ext_feats) #filter width is 5
     QAModel.save(net, args.dataset_folder, args.model_fname)
+
+    torch.set_num_threads(args.num_threads)
     
-    trainer = Trainer(net)
+    trainer = Trainer(net, args.eta, args.mom)
 
     best_accuracy = 0.0
     best_model = 0
     
     for i in range(args.epochs):
         logger.info('Training epoch {} -------------'.format(i+1))        
-        train_accuracy = trainer.train(args.dataset_folder, 'train', args.batch_size, cache_file)        
-        # sys.exit(0)
+        train_accuracy = trainer.train(args.dataset_folder, 'train', args.batch_size, cache_file, args.debugSingleBatch)        
+        if args.debugSingleBatch: sys.exit(0)
         dev_accuracy, dev_scores = trainer.test(args.dataset_folder, 'clean-dev', args.batch_size, cache_file)
         if dev_accuracy > best_accuracy:            
             best_model = i

sm-model/model.py

@@ -12,7 +12,7 @@ logger = logging.getLogger(__name__)
 logger.setLevel(logging.INFO)
 
 ch = logging.StreamHandler()
-ch.setLevel(logging.INFO)
+ch.setLevel(logging.DEBUG)
 formatter = logging.Formatter('%(levelname)s - %(message)s')
 ch.setFormatter(formatter)
 logger.addHandler(ch)
@@ -28,10 +28,12 @@ class QAModel(nn.Module):
     def load(in_folder, model_fname):
         return torch.load(os.path.join(in_folder, model_fname))
 
-    def __init__(self, input_n_dim, filter_width, ext_feats_size=4, n_classes=2):
+    def __init__(self, input_n_dim, filter_width, conv_filters=100, no_ext_feats=False, ext_feats_size=4, n_classes=2):
         super(QAModel, self).__init__()
 
-        self.conv_channels = 100
+        self.no_ext_feats = no_ext_feats
+
+        self.conv_channels = conv_filters
         n_hidden = 2*self.conv_channels + 1
 
         self.conv_q = nn.Sequential(
@@ -44,7 +46,7 @@ class QAModel(nn.Module):
             nn.Tanh()            
         )
         
-        self.combined_feature_vector = nn.Linear(2*self.conv_channels+ext_feats_size, n_hidden)
+        self.combined_feature_vector = nn.Linear(2*self.conv_channels + (0 if no_ext_feats else ext_feats_size), n_hidden)
         #TODO: add +1 to Linear layer^. Will need change in forward function
         self.combined_features_activation = nn.Tanh()
         self.dropout = nn.Dropout(0.5)
@@ -63,8 +65,15 @@ class QAModel(nn.Module):
         a = F.max_pool1d(a, a.size()[2])
         a = a.view(-1, self.conv_channels)
 
-        x = torch.cat([q, a, ext_feats], 1)        
-        # logger.debug('featvec x: {}'.format(x))
+        x = None
+        if self.no_ext_feats:
+            x = torch.cat([q, a], 1)        
+            logger.debug('no_ext_feats')
+        else:
+            x = torch.cat([q, a, ext_feats], 1)        
+            logger.debug('with ext_feats')
+        
+        logger.debug('featvec x: {}'.format(x))
         # logger.debug(x.creator)
 
         x = self.combined_feature_vector.forward(x)
@@ -73,9 +82,6 @@ class QAModel(nn.Module):
         x = self.hidden(x)
         x = self.logsoftmax(x)
 
-        logger.debug('x data {}'.format(x.data))
-        logger.debug('x grad {}'.format(x.grad))
-
         return x
 
 

sm-model/train.py

@@ -27,11 +27,11 @@ logger.addHandler(ch)
 
 class Trainer(object):
     
-    def __init__(self, model):
+    def __init__(self, model, eta, mom):
         self.reg = 1e-5
         self.model = model
         self.criterion = nn.CrossEntropyLoss()
-        self.optimizer = optim.SGD(self.model.parameters(), lr=0.001, weight_decay=self.reg)
+        self.optimizer = optim.SGD(self.model.parameters(), lr=eta, momentum=mom, weight_decay=self.reg)
 
 
     def regularize_loss(self, loss):       
@@ -150,7 +150,7 @@ class Trainer(object):
         return float(total_correct)/len(labels), y_pred
 
 
-    def train(self, dataset_folder, set_folder, batch_size, word_vectors_cache_file):
+    def train(self, dataset_folder, set_folder, batch_size, word_vectors_cache_file, debugSingleBatch):
         # read in training data
         questions, sentences, labels, vocab, maxlen_q, maxlen_s, ext_feats = \
                 utils.read_in_dataset(dataset_folder, set_folder)
@@ -186,7 +186,7 @@ class Trainer(object):
             # logger.debug('batch_loss {}, batch_correct {}'.format(batch_loss, batch_correct))
             train_loss += batch_loss
             train_correct += batch_correct
-            # break
+            if debugSingleBatch: break
 
         logger.info('train_correct {}'.format(train_correct))
         logger.info('train_loss {}'.format(train_loss))