def __init__(self, input_size, hidden_size, output_size):
# init para
weight_init_std = 0.01
self.params = {}
self.params['W1'] = weight_init_std * np.random.randn(
input_size, hidden_size)
self.params['b1'] = np.zeros(hidden_size)
self.params['gamma1'] = np.ones(hidden_size)
self.params['beta1'] = np.zeros(hidden_size)
self.params['W2'] = weight_init_std * np.random.randn(
hidden_size, output_size)
self.params['b2'] = np.zeros(output_size)
# create layers
self.layers = OrderedDict()
self.layers['Affine1'] = layers.Affine(self.params['W1'],
self.params['b1'])
self.layers['BatchNorm1'] = layers.BatchNormalization(
self.params['gamma1'], self.params['beta1'])
self.layers['Relu1'] = layers.Relu()
self.layers['Affine2'] = layers.Affine(self.params['W2'],
self.params['b2'])
self.lossLayer = layers.SoftmaxCrossEntropy(class_num=10)
self.loss_list = []
def __init__(self,
input_size,
hidden_size_list,
output_size,
activation='relu',
weight_init_std='relu',
weight_decay_lambda=0,
use_dropout=False,
dropout_ratio=0.5,
use_batchnorm=False):
self.input_size = input_size
self.output_size = output_size
self.hidden_size_list = hidden_size_list
self.hidden_layer_num = len(hidden_size_list)
self.use_dropout = use_dropout
self.weight_decay_lambda = weight_decay_lambda
self.use_batchnorm = use_batchnorm
self.params = {}
# 重みの初期化
self.__init_weight(weight_init_std)
# レイヤの生成
activation_layer = {'sigmoid': layers.Sigmoid, 'relu': layers.Relu}
self.layers = OrderedDict()
for idx in range(1, self.hidden_layer_num + 1):
self.layers['Affine' + str(idx)] = layers.Affine(
self.params['W' + str(idx)], self.params['b' + str(idx)])
if self.use_batchnorm:
self.params['gamma' + str(idx)] = np.ones(
hidden_size_list[idx - 1])
self.params['beta' + str(idx)] = np.zeros(
hidden_size_list[idx - 1])
self.layers['BatchNorm' +
str(idx)] = layers.BatchNormalization(
self.params['gamma' + str(idx)],
self.params['beta' + str(idx)])
self.layers['Activation_function' +
str(idx)] = activation_layer[activation]()
if self.use_dropout:
self.layers['Dropout' +
str(idx)] = layers.Dropout(dropout_ratio)
idx = self.hidden_layer_num + 1
self.layers['Affine' + str(idx)] = layers.Affine(
self.params['W' + str(idx)], self.params['b' + str(idx)])
self.last_layer = layers.SoftmaxWithLoss()
def __init__(self,
input_size,
hidden_size_list,
output_size,
activation='relu',
weight_init_std='relu',
weight_decay_lambda=0.0,
use_dropout=False,
dropout_ratio=0.5,
use_batchnorm=False):
self.input_size = input_size
self.output_size = output_size
self.hidden_size_list = hidden_size_list
self.hidden_layer_num = len(hidden_size_list)
self.weight_decay_lambda = weight_decay_lambda
self.use_dropout = use_dropout
self.use_batchnorm = use_batchnorm
self.params = {}
# weights initialization
self.__init_weight(weight_init_std)
# generate layers
activation_layer = {'sigmoid': layers.Sigmoid, 'relu': layers.Relu}
self.layers = OrderedDict()
for idx in range(1, self.hidden_layer_num + 1):
self.layers['Affine' + str(idx)] = layers.Affine(
self.params['W' + str(idx)], self.params['b' + str(idx)])
if self.use_batchnorm:
self.layers['BatchNorm' +
str(idx)] = layers.BatchNormalization(
self.params['gamma' + str(idx)],
self.params['beta' + str(idx)])
self.layers['Activation' +
str(idx)] = activation_layer[activation]()
if self.use_dropout:
self.layers['Dropout' +
str(idx)] = layers.Dropout(dropout_ratio)
# last Affine layer need no Activation & Batch Norm
idx = self.hidden_layer_num + 1
self.layers['Affine' + str(idx)] = layers.Affine(
self.params['W' + str(idx)], self.params['b' + str(idx)])
# self.last_layer = layers.SoftmaxCrossEntropy()
self.last_layer = layers.MSE()
# dict to save activation layer output
self.activation_dict = OrderedDict()
def __init__(self,
input_size,
hidden_size_list,
output_size,
activation='relu',
weight_init_std='relu',
weight_decay_lambda=0,
use_dropout=False,
dropout_ration=0.5,
use_batchnorm=False):
self.input_size = input_size
self.hidden_size_list = hidden_size_list
self.hidden_layer_num = len(hidden_size_list)
self.output_size = output_size
self.weight_decay_lambda = weight_decay_lambda
self.use_dropout = use_dropout
self.use_batchnorm = use_batchnorm
self.params = {}
# 가중치 초기화
self.__init_weight(weight_init_std)
# 계층 생성
activation_layer = {'sigmoid': layers.Sigmoid, 'relu': layers.Relu}
self.layers = OrderedDict()
# < 은닉층 생성 >
# self.hidden_layer_num 개수만큼
for idx in range(1, self.hidden_layer_num + 1):
# (1) Affine 계층
self.layers['Affine' + str(idx)] = layers.Affine(
self.params['W' + str(idx)], self.params['b' + str(idx)])
# (2) BatchNormalization 계층
if self.use_batchnorm:
# 각 계층별 배치 정규화 계층에서 사용할 매개변수 최기화
# 원본 그대로에서 시작하는 것으로 초기화. 1배 확대(gamma), 이동 0(beta)
self.params['gamma' + str(idx)] = np.ones(
hidden_size_list[idx - 1]) # 1
self.params['beta' + str(idx)] = np.zeros(
hidden_size_list[idx - 1]) # 0
self.layers['BatchNorm' +
str(idx)] = layers.BatchNormalization(
self.params['gamma' + str(idx)],
self.params['beta' + str(idx)])
# (3) 활성화 함수
self.layers['Activation_function' +
str(idx)] = activation_layer[activation]()
# (4) Dropout 계층
if self.use_dropout:
self.layers['Dropout' +
str(idx)] = layers.Dropout(dropout_ration)
# < 출력층 Affine 생성 >
idx = self.hidden_layer_num + 1
self.layers['Affine' + str(idx)] = layers.Affine(
self.params['W' + str(idx)], self.params['b' + str(idx)])
# < 출력층 생성 >
# 출력층 활성화 함수로 Softmax, 손실함수로 cross_entropy_error 사용
self.last_layer = layers.SoftmaxWithLoss()