Keras构建DNN模型的两种方式

1. 方式一

from keras.layers.core import Dense, Dropout
from keras.layers.normalization import BatchNormalization
from keras.models import Sequential,Model
from keras.layers import merge,Input,Add, LeakyReLU

class DNN(object):
    """
    Define a DNN model for classification.
    """

    def __init__(self, batch_size=128):
        self.batch_size = batch_size

    def build_model(self, input_dim, output_dim, hidden_dim_list=[128, 50]):
        '''
        :param inputdim: int type, the dim of input data.
        :param outputdim: int type, the number of class.
        '''
        input_1 = Input(shape=(input_dim,))
        hidden_0 = Dense(units=hidden_dim_list[0],activation='linear')(input_1)
        hidden_0 = LeakyReLU()(hidden_0)
        hidden_0 = BatchNormalization()(hidden_0)
        hidden_0 = Dropout(0.5)(hidden_0)
        #
        hidden_1 = Dense(hidden_dim_list[1], activation='linear')(hidden_0)
        hidden_1 = LeakyReLU()(hidden_1)
        hidden_1 = BatchNormalization()(hidden_1)
        hidden_1 = Dropout(0.5)(hidden_1)
        hidden_2 = Add()([hidden_0, hidden_1])
        hidden_3 = Dense(hidden_dim_list[2], activation='relu')(hidden_2)
        predictions = Dense(output_dim, activation='softmax')(hidden_3)
        model = Model(inputs=input_1, outputs=predictions)

        return model

2. 方式二

from keras.layers.core import Dense, Dropout
from keras.layers.normalization import BatchNormalization
from keras.models import Sequential

class DNN(object):
	"""
	Define a DNN model for classification.
	"""

	def __init__(self, batch_size=128):
		self.batch_size = batch_size

	def build_model(self, input_dim, output_dim, hidden_dim_list=[128, 50]):
		'''
		:param inputdim: int type, the dim of input data.
		:param outputdim: int type, the number of class.
		'''
		model = Sequential()
		model.add(Dense(hidden_dim_list[0], input_dim=input_dim, activation='relu'))
		model.add(BatchNormalization())
		model.add(Dropout(0.5))
		for i in range(1, len(hidden_dim_list)):
			model.add(Dense(hidden_dim_list[i], activation='relu'))
			model.add(BatchNormalization())
			model.add(Dropout(0.5))
		model.add(Dense(output_dim, activation='softmax'))

		return model