5.keras-Dropout剪枝操做的應用

keras-Dropout剪枝操做的應用

1.載入數據以及預處理

import numpy as np
from keras.datasets import mnist
from keras.utils import np_utils
from keras.models import Sequential
from keras.layers import *
from keras.optimizers import SGD

import os

import tensorflow as tf

# 載入數據
(x_train,y_train),(x_test,y_test) = mnist.load_data()

# 預處理
# 將（60000，28，28）轉化爲（600000，784），好輸入展開層
x_train = x_train.reshape(x_train.shape[0],-1)/255.0
x_test= x_test.reshape(x_test.shape[0],-1)/255.0
# 將輸出轉化爲one_hot編碼
y_train = np_utils.to_categorical(y_train,num_classes=10)
y_test = np_utils.to_categorical(y_test,num_classes=10)

2.建立網絡打印訓練結果

# 建立網絡
model = Sequential([

　　Dense(units=128,input_dim=784,bias_initializer='one',activation='tanh'),
　　# Dropout進行減枝，使得部分訓練參數失效，避免過擬和
　　Dropout(0.4),
　　Dense(units=128,bias_initializer='one',activation='tanh'),
　　Dropout(0.4),
　　Dense(units=10,bias_initializer='one',activation='softmax')

]) 
# 編譯 
# 自定義優化器 
sgd = SGD(lr=0.1) model.compile(optimizer=sgd,
　　　　　　　　
　　　　　　　　# 運用交叉熵
              loss='categorical_crossentropy',
              metrics=['accuracy'])

model.fit(x_train,y_train,batch_size=32,epochs=10,validation_split=0.2)

# 評估模型
loss,acc = model.evaluate(x_test,y_test,)
print('\ntest loss',loss)
print('test acc',acc)

out:

Epoch 1/10

32/48000 [..............................] - ETA: 5:04 - loss: 2.7763 - acc: 0.1250
576/48000 [..............................] - ETA: 21s - loss: 2.6202 - acc: 0.1354

......

......

Epoch 10/10

47744/48000 [============================>.] - ETA: 0s - loss: 0.1830 - acc: 0.9448
48000/48000 [==============================] - 3s 72us/step - loss: 0.1831 - acc: 0.9449 - val_loss: 0.1210 - val_acc: 0.9649

 

32/10000 [..............................] - ETA: 0s
1824/10000 [====>.........................] - ETA: 0s
3616/10000 [=========>....................] - ETA: 0s
5472/10000 [===============>..............] - ETA: 0s
7456/10000 [=====================>........] - ETA: 0s
9440/10000 [===========================>..] - ETA: 0s
10000/10000 [==============================] - 0s 27us/step

test loss 0.11740412595644593test acc 0.9652