Enable model saving during training

fluid/DeepASR/train.py

@@ -95,6 +95,11 @@ def parse_args():
         type=str,
         default='data/val_label.lst',
         help='label list path for validation.')
+    parser.add_argument(
+        '--model_save_dir',
+        type=str,
+        default='./checkpoints',
+        help='directory to save model.')
     args = parser.parse_args()
     return args
 
@@ -109,8 +114,8 @@ def print_arguments(args):
 def train(args):
     """train in loop."""
 
-    _, avg_cost, accuracy = stacked_lstmp_model(args.hidden_dim, args.proj_dim,
-                                                args.stacked_num, args.parallel)
+    prediction, avg_cost, accuracy = stacked_lstmp_model(
+        args.hidden_dim, args.proj_dim, args.stacked_num, args.parallel)
 
     adam_optimizer = fluid.optimizer.Adam(learning_rate=args.learning_rate)
     adam_optimizer.minimize(avg_cost)
@@ -192,10 +197,16 @@ def train(args):
             else:
                 sys.stdout.write('.')
                 sys.stdout.flush()
-
+        # run test
         val_cost, val_acc = test(exe)
         pass_end_time = time.time()
         time_consumed = pass_end_time - pass_start_time
+        # save model 
+        if args.model_save_dir is not None:
+            model_path = os.path.join(
+                args.model_save_dir, "deep_asr.pass_" + str(pass_id) + ".model")
+            fluid.io.save_inference_model(model_path, ["feature"],
+                                          [prediction], exe)
 
         print("\nPass %d, time consumed: %f s, val cost: %f, val acc: %f\n" %
               (pass_id, time_consumed, val_cost, val_acc))
@@ -205,4 +216,8 @@ if __name__ == '__main__':
     args = parse_args()
     print_arguments(args)
 
+    if args.model_save_dir is not None and \
+       not os.path.exists(args.model_save_dir):
+        os.mkdir(args.model_save_dir)
+
     train(args)