41 def calculate_model_size(model):  argument
42     print(model.summary())
45             for v in model.trainable_variables
47     print("Model size:", sum(var_sizes) / 1024, "KB")
52     model = tf.keras.Sequential([
73     model.load_weights("./netmodels/CNN/weights.h5")
74     return model, model_path
79     model = tf.keras.Sequential([
89     return model, model_path
101     if args.model == "CNN":
102         model, model_path = build_cnn(seq_length)
103     elif args.model == "LSTM":
104         model, model_path = build_lstm(seq_length)
106         print("Please input correct model name.(CNN    LSTM)")
107     return model, model_path
111         model,  argument
120     """Trains the model."""
121     calculate_model_size(model)
124     model.compile(
140     model.fit(
147     loss, acc = model.evaluate(test_data)
148     pred = np.argmax(model.predict(test_data), axis=1)
155     # Convert the model to the TensorFlow Lite format without quantization
156     converter = tf.lite.TFLiteConverter.from_keras_model(model)
159     # Save the model to disk
160     open("model.tflite", "wb").write(tflite_model)
162     # Convert the model to the TensorFlow Lite format with quantization
163     converter = tf.lite.TFLiteConverter.from_keras_model(model)
167     # Save the model to disk
170     basic_model_size = os.path.getsize("model.tflite")
171     print("Basic model is %d bytes" % basic_model_size)
173     print("Quantized model is %d bytes" % quantized_model_size)
180     parser.add_argument("--model", "-m")
196     model, model_path = build_net(args, seq_length)  variable
199     train_net(model, model_path, train_len, train_data, valid_len, valid_data,
200                         test_len, test_data, args.model)