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

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

 

 

目录

输出结果

设计思路

核心代码

更多输出

---

 

 

 

输出结果

 

设计思路

190417更新

 

 

 

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

 

 

更多输出

1、DNN[6*100+ReLU,SGD]: accuracy of not dropout on Minist dataset

train loss:2.3364575765992637
=== epoch:1, train acc:0.10333333333333333, test acc:0.1088 ===
train loss:2.414526554119518
train loss:2.341182306768928
train loss:2.3072782723352496
=== epoch:2, train acc:0.09666666666666666, test acc:0.1103 ===
train loss:2.2600377181768887
train loss:2.263350960525319
train loss:2.2708260374887645

……

=== epoch:298, train acc:1.0, test acc:0.7709 ===
train loss:0.00755416896470134
train loss:0.009934657874546435
train loss:0.008421672959852643
=== epoch:299, train acc:1.0, test acc:0.7712 ===
train loss:0.007142981215285884
train loss:0.008205245499586114
train loss:0.007319626293763803
=== epoch:300, train acc:1.0, test acc:0.7707 ===
train loss:0.00752230499930163
train loss:0.008431046288276818
train loss:0.008067532729014863
=== epoch:301, train acc:1.0, test acc:0.7707 ===
train loss:0.010729407851274233
train loss:0.007776889701033221
=============== Final Test Accuracy ===============
test acc:0.771

 

2、DNN[6*100+ReLU,SGD]: accuracy of dropout(0.2) on Minist dataset

train loss:2.3064018541384437
=== epoch:1, train acc:0.11, test acc:0.1112 ===
train loss:2.316626942558816
train loss:2.314434337198633
train loss:2.318862771955365
=== epoch:2, train acc:0.11333333333333333, test acc:0.1128 ===
train loss:2.3241989320140717
train loss:2.317694982413387
train loss:2.3079716553885006

……

=== epoch:298, train acc:0.6266666666666667, test acc:0.5168 ===
train loss:1.2359381134877185
train loss:1.2833380447791383
train loss:1.2728131428100005
=== epoch:299, train acc:0.63, test acc:0.52 ===
train loss:1.1687601000183936
train loss:1.1435412548991142
train loss:1.3854277174616834
=== epoch:300, train acc:0.6333333333333333, test acc:0.5244 ===
train loss:1.3039470016588997
train loss:1.2359979876607923
train loss:1.2871396654831204
=== epoch:301, train acc:0.63, test acc:0.5257 ===
train loss:1.1690084424502523
train loss:1.1820777530873694
=============== Final Test Accuracy ===============
test acc:0.5269

 

相关文章
CSDN：2019.04.09起