vdcnn.py 文件源码

def build_model(num_filters, num_classes, sequence_max_length=512, num_quantized_chars=71, embedding_size=16, learning_rate=0.001, top_k=3, model_path=None):

    inputs = Input(shape=(sequence_max_length, ), dtype='int32', name='inputs')

    embedded_sent = Embedding(num_quantized_chars, embedding_size, input_length=sequence_max_length)(inputs)

    # First conv layer
    conv = Conv1D(filters=64, kernel_size=3, strides=2, padding="same")(embedded_sent)

    # Each ConvBlock with one MaxPooling Layer
    for i in range(len(num_filters)):
        conv = ConvBlockLayer(get_conv_shape(conv), num_filters[i])(conv)
        conv = MaxPooling1D(pool_size=3, strides=2, padding="same")(conv)

    # k-max pooling (Finds values and indices of the k largest entries for the last dimension)
    def _top_k(x):
        x = tf.transpose(x, [0, 2, 1])
        k_max = tf.nn.top_k(x, k=top_k)
        return tf.reshape(k_max[0], (-1, num_filters[-1] * top_k))
    k_max = Lambda(_top_k, output_shape=(num_filters[-1] * top_k,))(conv)

    # 3 fully-connected layer with dropout regularization
    fc1 = Dropout(0.2)(Dense(512, activation='relu', kernel_initializer='he_normal')(k_max))
    fc2 = Dropout(0.2)(Dense(512, activation='relu', kernel_initializer='he_normal')(fc1))
    fc3 = Dense(num_classes, activation='softmax')(fc2)

    # define optimizer
    sgd = SGD(lr=learning_rate, decay=1e-6, momentum=0.9, nesterov=False)
    model = Model(inputs=inputs, outputs=fc3)
    model.compile(optimizer=sgd, loss='mean_squared_error', metrics=['accuracy'])

    if model_path is not None:
        model.load_weights(model_path)

    return model