前一篇文章tensorflow入門應(yīng)用方法——線性回歸和邏輯回歸中，主要闡述了應(yīng)用tensorflow搭建線性回歸和邏輯回歸的相關(guān)代碼刚盈。其中胎许，線性回歸和邏輯回歸模型是一個(gè)簡單的單層網(wǎng)絡(luò)找前。因此丈屹，本文較為深入闡述應(yīng)用tensorflow搭建擁有兩層隱藏層的網(wǎng)絡(luò)方法流程调俘。

具體步驟

1. 初始化每層的神經(jīng)元數(shù)量 需要定義出每層的輸出神經(jīng)元數(shù)量
2. 輸入輸出初始化 一般用placeholder占位符定義x,y
3. 網(wǎng)絡(luò)參數(shù)初始化 w權(quán)值和bias隨機(jī)初始化范圍(-1,1)，bias也可初始化為0
4. 前向計(jì)算 需要選擇激活函數(shù)
5. 損失函數(shù)和最優(yōu)化定義 可以選擇不同的損失函數(shù)和優(yōu)化算法
6. 預(yù)測精度函數(shù)定義
7. 初始化所有變量 init = tf.global_variables_initializer()
8. 迭代運(yùn)算

深度網(wǎng)絡(luò)代碼實(shí)現(xiàn)

首先加載mnist數(shù)據(jù)集用于模型訓(xùn)練

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data


# 加載數(shù)據(jù)
mnist = input_data.read_data_sets('data/', one_hot=True)

然后旺垒，開始搭建深度模型具體步驟

# 神經(jīng)元個(gè)數(shù)初始化
n_hidden_1 = 256
n_hidden_2 = 128
n_input = 784  # 28*28*1的輸入
n_class = 10
# 輸入和輸出
x = tf.placeholder('float', [None, n_input])
y = tf.placeholder('float', [None, n_class])


# 網(wǎng)絡(luò)參數(shù)初始化
stddev = 0.1  # 指定高斯分布方差
weights = {
    'w1': tf.Variable(tf.random_normal([n_input, n_hidden_1], stddev=stddev)),
    'w2': tf.Variable(tf.random_normal([n_hidden_1, n_hidden_2], stddev=stddev)),
    'out': tf.Variable(tf.random_normal([n_hidden_2, n_class], stddev=stddev))
}

biases = {
    'b1': tf.Variable(tf.random_normal([n_hidden_1])),
    'b2': tf.Variable(tf.random_normal([n_hidden_2])),
    'out': tf.Variable(tf.random_normal([n_class]))
}

print('network ready')


# 訓(xùn)練網(wǎng)絡(luò)

# 前向計(jì)算
def multilayer_perception(_X, _weight, _biases):
    layer1 = tf.nn.sigmoid(tf.add(tf.matmul(_X, _weight['w1']), _biases['b1']))
    layer2 = tf.nn.sigmoid(tf.add(tf.matmul(layer1, _weight['w2']), _biases['b2']))
    return (tf.matmul(layer2, _weight['out']) +_biases['out'])

pred = multilayer_perception(x, weights, biases)


# 損失函數(shù)和最優(yōu)化
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pred, labels=y))
optm = tf.train.GradientDescentOptimizer(learning_rate=0.01).minimize(cost)


# 計(jì)算精度
corr = tf.equal(tf.arg_max(pred, 1), tf.arg_max(y, 1))
accr = tf.reduce_mean(tf.cast(corr, 'float'))


# 初始化操作
init = tf.global_variables_initializer()
print('function ready')


# 迭代計(jì)算
training_epochs = 20
batch_size = 100
display_step = 4
sess = tf.Session()
sess.run(init)

for epoch in range(training_epochs):
    avg_cost = 0
    total_batch = int(mnist.train.num_examples/batch_size)
    for i in range(total_batch):
        batch_xs, batch_ys = mnist.train.next_batch(batch_size)
        feeds = {x: batch_xs, y: batch_ys}
        sess.run(optm, feed_dict=feeds)
        avg_cost += sess.run(cost, feed_dict=feeds)/total_batch
    # 顯示結(jié)果
    if (epoch+1) % display_step == 0:
        print('Epoch: %03d/%03d cost: %.9f' % (epoch+1, training_epochs, avg_cost))
        feeds = {x: batch_xs, y: batch_ys}
        train_acc = sess.run(accr, feed_dict=feeds)
        print('Train accuracy: %.3f' % train_acc)
        feeds = {x: mnist.test.images, y: mnist.test.labels}
        test_acc = sess.run(accr, feed_dict=feeds)
        print('Test accuracy: %.3f' % test_acc)

print('optmization finished')

訓(xùn)練模型時(shí)主要使用梯度下降算法彩库，迭代20次，每4步顯示一次結(jié)果：

Epoch: 004/020 cost: 1.835647699
Train accuracy: 0.720
Test accuracy: 0.682
Epoch: 008/020 cost: 0.946993933
Train accuracy: 0.780
Test accuracy: 0.804
Epoch: 012/020 cost: 0.642496496
Train accuracy: 0.890
Test accuracy: 0.849
Epoch: 016/020 cost: 0.519645967
Train accuracy: 0.800
Test accuracy: 0.869
Epoch: 020/020 cost: 0.452667181
Train accuracy: 0.880
Test accuracy: 0.885
optmization finished

總結(jié)

在本文中先蒋，經(jīng)過20次的迭代訓(xùn)練骇钦，模型在訓(xùn)練集上的精度為88.0%，測試集上的精度為88.5%竞漾。