DL之DNN：利用MultiLayerNetExtend模型【6*100+ReLU+SGD,dropout】对Mnist数据集训练来抑制过拟合

输出结果

设计思路

核心代码

class RMSprop:

   def __init__(self, lr=0.01, decay_rate = 0.99):

       self.lr = lr

       self.decay_rate = decay_rate

       self.h = None

   def update(self, params, grads):

       if self.h is None:

           self.h = {}

           for key, val in params.items():

               self.h[key] = np.zeros_like(val)

       for key in params.keys():

           self.h[key] *= self.decay_rate

           self.h[key] += (1 - self.decay_rate) * grads[key] * grads[key]

           params[key] -= self.lr * grads[key] / (np.sqrt(self.h[key]) + 1e-7)

class Nesterov:

   def __init__(self, lr=0.01, momentum=0.9):

       self.lr = lr

       self.momentum = momentum

       self.v = None

   def update(self, params, grads):

       if self.v is None:

           self.v = {}

           for key, val in params.items():

               self.v[key] = np.zeros_like(val)

       for key in params.keys():

           self.v[key] *= self.momentum

           self.v[key] -= self.lr * grads[key]

           params[key] += self.momentum * self.momentum * self.v[key]

           params[key] -= (1 + self.momentum) * self.lr * grads[key]

use_dropout = True  

dropout_ratio = 0.2

network = MultiLayerNetExtend(input_size=784, hidden_size_list=[100, 100, 100, 100, 100, 100],

                             output_size=10, use_dropout=use_dropout, dropout_ration=dropout_ratio)

trainer = Trainer(network, x_train, t_train, x_test, t_test, epochs=301, mini_batch_size=100,

                 optimizer='sgd', optimizer_param={'lr': 0.01}, verbose=True)  

trainer.train()                                                                

train_acc_list, test_acc_list = trainer.train_acc_list, trainer.test_acc_list