Validation While trainnig

caption_generator/caption_generator.py

@@ -19,6 +19,7 @@ def __init__(self):
         self.index_word = None
         self.word_index = None
         self.total_samples = None
+        self.total_samples_dev = None
         self.encoded_images = pickle.load( open( "encoded_images.p", "rb" ) )
         self.variable_initializer()
 
@@ -34,8 +35,20 @@ def variable_initializer(self):
         self.total_samples=0
         for text in caps:
             self.total_samples+=len(text.split())-1
-        print "Total samples : "+str(self.total_samples)
-
+        print "Total samples Train : "+str(self.total_samples)
+
+        df_dev = pd.read_csv('Flickr8k_text/flickr_8k_dev_dataset.txt', delimiter='\t')
+        nb_samples = df_dev.shape[0]
+        iter = df_dev.iterrows()
+        for i in range(nb_samples):
+            x = iter.next()
+            caps.append(x[1][1])
+
+        self.total_samples_dev=0
+        for text in caps:
+            self.total_samples_dev+=len(text.split())-1
+        print "Total samples Dev : "+str(self.total_samples_dev)
+
         words = [txt.split() for txt in caps]
         unique = []
         for word in words:
@@ -59,13 +72,13 @@ def variable_initializer(self):
         print "Variables initialization done!"
 
 
-    def data_generator(self, batch_size = 32):
+    def data_generator(self, path, batch_size = 32):
         partial_caps = []
         next_words = []
         images = []
         print "Generating data..."
         gen_count = 0
-        df = pd.read_csv('Flickr8k_text/flickr_8k_train_dataset.txt', delimiter='\t')
+        df = pd.read_csv(path, delimiter='\t')
         nb_samples = df.shape[0]
         iter = df.iterrows()
         caps = []
@@ -85,6 +98,7 @@ def data_generator(self, batch_size = 32):
                 for i in range(len(text.split())-1):
                     total_count+=1
                     partial = [self.word_index[txt] for txt in text.split()[:i+1]]
+                    #print partial, [txt for txt in text.split()[:i+1]]
                     partial_caps.append(partial)
                     next = np.zeros(self.vocab_size)
                     next[self.word_index[text.split()[i+1]]] = 1
@@ -139,4 +153,5 @@ def create_model(self, ret_model = False):
         return model
 
     def get_word(self,index):
-        return self.index_word[index]
+        return self.index_word[index]
+

caption_generator/prepare_dataset.py

@@ -36,12 +36,19 @@ def prepare_dataset(no_imgs = -1):
 	test_imgs = f_test_images.read().strip().split('\n') if no_imgs == -1 else f_test_images.read().strip().split('\n')[:no_imgs]
 	f_test_images.close()
 
+	f_dev_images = open('Flickr8k_text/Flickr_8k.devImages.txt','rb')
+	dev_imgs = f_dev_images.read().strip().split('\n') if no_imgs == -1 else f_dev_images.read().strip().split('\n')[:no_imgs]
+	f_dev_images.close()
+
 	f_train_dataset = open('Flickr8k_text/flickr_8k_train_dataset.txt','wb')
 	f_train_dataset.write("image_id\tcaptions\n")
 
 	f_test_dataset = open('Flickr8k_text/flickr_8k_test_dataset.txt','wb')
 	f_test_dataset.write("image_id\tcaptions\n")
 
+	f_dev_dataset = open('Flickr8k_text/flickr_8k_dev_dataset.txt','wb')
+	f_dev_dataset.write("image_id\tcaptions\n")
+
 	f_captions = open('Flickr8k_text/Flickr8k.token.txt', 'rb')
 	captions = f_captions.read().strip().split('\n')
 	data = {}
@@ -76,6 +83,17 @@ def prepare_dataset(no_imgs = -1):
 			f_test_dataset.flush()
 			c_test += 1
 	f_test_dataset.close()
+
+	c_dev = 0
+	for img in dev_imgs:
+		encoded_images[img] = get_encoding(encoding_model, img)
+		for capt in data[img]:
+			caption = "<start> "+capt+" <end>"
+			f_dev_dataset.write(img+"\t"+caption+"\n")
+			f_dev_dataset.flush()
+			c_dev += 1
+	f_dev_dataset.close()
+
 	with open( "encoded_images.p", "wb" ) as pickle_f:
 		pickle.dump( encoded_images, pickle_f )  
 	return [c_train, c_test]
@@ -84,3 +102,4 @@ def prepare_dataset(no_imgs = -1):
 	c_train, c_test = prepare_dataset()
 	print "Training samples = "+str(c_train)
 	print "Test samples = "+str(c_test)
+

caption_generator/train_model.py

@@ -10,10 +10,10 @@ def train_model(weight = None, batch_size=32, epochs = 10):
         model.load_weights(weight)
 
     counter = 0
-    file_name = 'weights-improvement-{epoch:02d}.hdf5'
-    checkpoint = ModelCheckpoint(file_name, monitor='loss', verbose=1, save_best_only=True, mode='min')
+    file_name = 'weights-improvement-epoch-{epoch:02d}-val_acc-{val_acc:.5f}-val_loss-{val_loss:.5f}.hdf5'
+    checkpoint = ModelCheckpoint(file_name, monitor='val_acc', verbose=1, save_best_only=True, mode='max')
     callbacks_list = [checkpoint]
-    model.fit_generator(cg.data_generator(batch_size=batch_size), steps_per_epoch=cg.total_samples/batch_size, epochs=epochs, verbose=2, callbacks=callbacks_list)
+    model.fit_generator(cg.data_generator(path = 'Flickr8k_text/flickr_8k_train_dataset.txt',batch_size=batch_size), steps_per_epoch=cg.total_samples/batch_size, epochs=epochs, verbose=2, callbacks=callbacks_list,validation_steps=cg.total_samples_dev/batch_size,validation_data=cg.data_generator(path = 'Flickr8k_text/flickr_8k_dev_dataset.txt',batch_size=batch_size))
     try:
         model.save('Models/WholeModel.h5', overwrite=True)
         model.save_weights('Models/Weights.h5',overwrite=True)
@@ -22,4 +22,5 @@ def train_model(weight = None, batch_size=32, epochs = 10):
     print "Training complete...\n"
 
 if __name__ == '__main__':
-    train_model(epochs=50)
+    train_model(epochs=100)
+