Update the senta demo (#56)

demo/senta/predict.py

@@ -15,8 +15,9 @@ import paddlehub as hub
 
 # yapf: disable
 parser = argparse.ArgumentParser(__doc__)
-parser.add_argument("--checkpoint_dir", type=str, default=None, help="Directory to model checkpoint")
-parser.add_argument("--use_gpu", type=ast.literal_eval, default=False, help="Whether use GPU for finetuning, input should be True or False")
+parser.add_argument("--checkpoint_dir", type=str,                 default=None, help="Directory to model checkpoint")
+parser.add_argument("--use_gpu",        type=ast.literal_eval,    default=True, help="Whether use GPU for finetuning, input should be True or False")
+parser.add_argument("--use_pyreader",   type=ast.literal_eval,    default=False, help="Whether use pyreader to feed data.")
 args = parser.parse_args()
 # yapf: enable.
 
@@ -30,42 +31,39 @@ if __name__ == '__main__':
     reader = hub.reader.LACClassifyReader(
         dataset=dataset, vocab_path=module.get_vocab_path())
 
-    place = fluid.CUDAPlace(0) if args.use_gpu else fluid.CPUPlace()
-    exe = fluid.Executor(place)
-    with fluid.program_guard(program):
-        sent_feature = outputs["sentence_feature"]
+    strategy = hub.AdamWeightDecayStrategy(
+        weight_decay=0.01,
+        warmup_proportion=0.1,
+        learning_rate=5e-5,
+        lr_scheduler="linear_decay",
+        optimizer_name="adam")
 
-        # Define a classfication finetune task by PaddleHub's API
-        cls_task = hub.create_text_cls_task(
-            feature=sent_feature, num_classes=dataset.num_labels)
+    config = hub.RunConfig(
+        use_data_parallel=False,
+        use_pyreader=args.use_pyreader,
+        use_cuda=args.use_gpu,
+        batch_size=1,
+        enable_memory_optim=False,
+        checkpoint_dir=args.checkpoint_dir,
+        strategy=strategy)
 
-        # Setup feed list for data feeder
-        # Must feed all the tensor of senta's module need
-        feed_list = [inputs["words"].name, cls_task.variable('label').name]
+    sent_feature = outputs["sentence_feature"]
 
-        # classificatin probability tensor
-        probs = cls_task.variable("probs")
+    feed_list = [inputs["words"].name]
 
-        pred = fluid.layers.argmax(probs, axis=1)
+    cls_task = hub.TextClassifierTask(
+        data_reader=reader,
+        feature=sent_feature,
+        feed_list=feed_list,
+        num_classes=dataset.num_labels,
+        config=config)
 
-        # load best model checkpoint
-        fluid.io.load_persistables(exe, args.checkpoint_dir)
+    data = ["这家餐厅很好吃", "这部电影真的很差劲"]
 
-        inference_program = program.clone(for_test=True)
-
-        data_feeder = fluid.DataFeeder(feed_list=feed_list, place=place)
-        test_reader = reader.data_generator(phase='test', shuffle=False)
-        test_examples = dataset.get_test_examples()
-        total = 0
-        correct = 0
-        for index, batch in enumerate(test_reader()):
-            pred_v = exe.run(
-                feed=data_feeder.feed(batch),
-                fetch_list=[pred.name],
-                program=inference_program)
-            total += 1
-            if (pred_v[0][0] == int(test_examples[index].label)):
-                correct += 1
-                acc = 1.0 * correct / total
-            print("%s\tpredict=%s" % (test_examples[index], pred_v[0][0]))
-    print("accuracy = %f" % acc)
+    results = cls_task.predict(data=data)
+    index = 0
+    for batch_result in results:
+        batch_result = np.argmax(batch_result, axis=2)[0]
+        for result in batch_result:
+            print("%s\tpredict=%s" % (data[index], result))
+            index += 1

demo/senta/run_finetune.sh

-export CUDA_VISIBLE_DEVICES=5
+export CUDA_VISIBLE_DEVICES=0
 
 DATASET="chnsenticorp"
 CKPT_DIR="./ckpt_${DATASET}"

demo/senta/senta_finetune.py

@@ -8,7 +8,7 @@ import paddlehub as hub
 # yapf: disable
 parser = argparse.ArgumentParser(__doc__)
 parser.add_argument("--num_epoch", type=int, default=3, help="Number of epoches for fine-tuning.")
-parser.add_argument("--use_gpu", type=ast.literal_eval, default=False, help="Whether use GPU for finetuning, input should be True or False")
+parser.add_argument("--use_gpu", type=ast.literal_eval, default=True, help="Whether use GPU for finetuning, input should be True or False")
 parser.add_argument("--checkpoint_dir", type=str, default=None, help="Directory to model checkpoint")
 parser.add_argument("--batch_size", type=int, default=32, help="Total examples' number in batch for training.")
 args = parser.parse_args()
@@ -19,7 +19,7 @@ if __name__ == '__main__':
     module = hub.Module(name="senta_bilstm")
     inputs, outputs, program = module.context(trainable=True)
 
-    # Step2: Download dataset and use TextClassificationReader to read dataset
+    # Step2: Download dataset and use LACClassifyReader to read dataset
     dataset = hub.dataset.ChnSentiCorp()
 
     reader = hub.reader.LACClassifyReader(
@@ -27,13 +27,9 @@ if __name__ == '__main__':
 
     sent_feature = outputs["sentence_feature"]
 
-    # Define a classfication finetune task by PaddleHub's API
-    cls_task = hub.create_text_cls_task(
-        feature=sent_feature, num_classes=dataset.num_labels)
-
     # Setup feed list for data feeder
     # Must feed all the tensor of senta's module need
-    feed_list = [inputs["words"].name, cls_task.variable('label').name]
+    feed_list = [inputs["words"].name]
 
     strategy = hub.finetune.strategy.AdamWeightDecayStrategy(
         learning_rate=1e-4, weight_decay=0.01, warmup_proportion=0.05)
@@ -45,7 +41,14 @@ if __name__ == '__main__':
         checkpoint_dir=args.checkpoint_dir,
         strategy=strategy)
 
+    # Define a classfication finetune task by PaddleHub's API
+    cls_task = hub.TextClassifierTask(
+        data_reader=reader,
+        feature=sent_feature,
+        feed_list=feed_list,
+        num_classes=dataset.num_labels,
+        config=config)
+
     # Finetune and evaluate by PaddleHub's API
     # will finish training, evaluation, testing, save model automatically
-    hub.finetune_and_eval(
-        task=cls_task, data_reader=reader, feed_list=feed_list, config=config)
+    cls_task.finetune_and_eval()