def build_mlp():
model = build_single_input()
model.add(Flatten())
model.add(Dropout(0.5))
model.add(Dense(embedding_size * 4, activation='relu', kernel_regularizer=l2(0.001)))
model.add(Dropout(0.5))
return build_single_output(model)
def build_fcn():
"""
全卷积神经网络
:param model:
:return:
"""
model = build_single_input()
return build_single_output(model)
def build_text_cnn():
"""
文本卷积神经网络
:param model:
:return:
"""
model = build_single_input()
return build_single_output(model)
def build_le_net():
"""
卷积神经网络
:param model:
:return:
"""
model = build_single_input()
model.add(Conv1D(embedding_size * 2, 2, activation='relu', kernel_regularizer=l2(0.001)))
model.add(MaxPooling1D())
model.add(Conv1D(embedding_size * 3, 2, activation='relu', kernel_regularizer=l2(0.001)))
model.add(MaxPooling1D())
model.add(Flatten())
model.add(Dropout(0.2))
model.add(Dense(embedding_size * 3, activation='relu', kernel_regularizer=l2(0.001)))
model.add(Dropout(0.2))
return build_single_output(model)
def build_nin():
"""
网络中的网络，训练这个数据效果不好，可能是参数不合适
:param model:
:return:
"""
model = build_single_input()
model.add(nin_block(embedding_size * 2, 3, 2, 'valid'))
model.add(MaxPooling1D())
model.add(nin_block(embedding_size * 3, 2, 1, 'same'))
model.add(MaxPooling1D())
model.add(nin_block(embedding_size * 4, 2, 1, 'same'))
model.add(MaxPooling1D())
model.add(Dropout(0.5))
model.add(nin_block(1, 2, 1, 'same'))
model.add(GlobalAveragePooling1D())
return build_single_output(model)
def build_conv1d():
model = build_single_input()
model.add(Conv1D(embedding_size * 2, 2, activation='relu', kernel_regularizer=l2(0.001)))
model.add(Conv1D(embedding_size * 2, 2, activation='relu', kernel_regularizer=l2(0.001)))
model.add(MaxPooling1D())
model.add(Conv1D(embedding_size * 4, 2, activation='relu', kernel_regularizer=l2(0.001)))
model.add(Conv1D(embedding_size * 4, 2, activation='relu', kernel_regularizer=l2(0.001)))
model.add(MaxPooling1D())
model.add(Conv1D(embedding_size * 8, 2, activation='relu', kernel_regularizer=l2(0.001)))
model.add(Conv1D(embedding_size * 8, 2, activation='relu', kernel_regularizer=l2(0.001)))
# model.add(MaxPooling1D())
# model.add(Conv1D(embedding_size * 16, 2, activation='relu', kernel_regularizer=l2(0.001)))
# model.add(Conv1D(embedding_size * 16, 2, activation='relu', kernel_regularizer=l2(0.001)))
# model.add(MaxPooling1
# model.add(Flatten())
# model.add(Conv1D(embedding_size * 2, 3, strides=2, activation='relu', kernel_regularizer=l2(0.001)))
# model.add(Conv1D(embedding_size * 4, 3, strides=2, activation='relu', kernel_regularizer=l2(0.001)))
# model.add(Conv1D(embedding_size * 8, 3, strides=2, activation='relu', kernel_regularizer=l2(0.001)))
# model.add(Conv1D(embedding_size * 16, 3, strides=2, activation='relu', kernel_regularizer=l2(0.001)))
model.add(GlobalMaxPooling1D())
model.add(Dropout(0.2))
return build_single_output(model)
def build_vgg():
"""
使用重复元素的网络
:param model:
:return:
"""
model = build_single_input()
conv_arch = ((2, 64), (2, 64), (2, 128), (2, 128))
for (num_convs, num_channels) in conv_arch:
model.add(vgg_block(num_convs, num_channels))
model.add(Flatten())
model.add(
Dense(embedding_size * 2,
activation='relu',
kernel_regularizer=l2(0.001)))
model.add(Dropout(0.5))
model.add(
Dense(embedding_size * 2,
activation='relu',
kernel_regularizer=l2(0.001)))
model.add(Dropout(0.5))
return build_single_output(model)
def build_alex_net():
"""
深度总面积神经网络
:param model:
:return:
"""
model = build_single_input()
model.add(Conv1D(embedding_size * 2, kernel_size=2, activation='relu', kernel_regularizer=l2(0.001)))
model.add(Conv1D(embedding_size * 2, kernel_size=2, activation='relu', kernel_regularizer=l2(0.001)))
model.add(MaxPooling1D())
model.add(Conv1D(embedding_size * 4, kernel_size=2, padding='same', activation='relu', kernel_regularizer=l2(0.001)))
model.add(Conv1D(embedding_size * 4, kernel_size=2, padding='same', activation='relu', kernel_regularizer=l2(0.001)))
model.add(MaxPooling1D())
model.add(Conv1D(embedding_size * 8, kernel_size=2, padding='valid', activation='relu', kernel_regularizer=l2(0.001)))
model.add(Conv1D(embedding_size * 8, kernel_size=2, padding='valid', activation='relu', kernel_regularizer=l2(0.001)))
model.add(Conv1D(embedding_size * 4, kernel_size=2, padding='valid', activation='relu', kernel_regularizer=l2(0.001)))
model.add(MaxPooling1D())
model.add(Flatten())
model.add(Dense(embedding_size * 8, activation='relu', kernel_regularizer=l2(0.001)))
model.add(Dropout(0.5))
model.add(Dense(embedding_size * 8, activation='relu', kernel_regularizer=l2(0.001)))
model.add(Dropout(0.5))
return build_single_output(model)
def build_global_max_pooling1d():
model = build_single_input()
model.add(GlobalMaxPooling1D())
return build_single_output(model)