TensorFlow神經網路：模組化的神經網路八股
阿新 • • 發佈：2018-11-14
1、前向傳播：
- 搭建從輸入到輸出的網路結構
forward.py:
# 定義前向傳播過程
def forward(x, regularizer):
w =
b =
y =
return y
# 給w賦初值，並把w的正則化損失加到總損失中
def get_weight(shape, regularizer):
w = tf.Variable()
tf.add_to_collection('losses', tf.contrib.layers.l2_regularizer(regularizer)(w))
return w
def get_bias 2、反向傳播
- 訓練網路，優化網路引數，提高模型準確性
backward.py：
# 定義反向傳播
def backward():
# 對資料集x和標準答案y_佔位
x = tf.placeholder()
y_ = tf.placeholder(
)
# 利用forward模組復現前向傳播網路的結構，計算得到y
y = forward.forward(x, REGULARIZER)
# 定義輪數計數器
global_step = tf.Variable(0, trainable = 3、判斷主檔案
# 判斷python執行檔案是否為主檔案，如果是，則執行
if __name__ == '__main__':
backward()
4、例項模組化展示
- 加入指數衰減學習率–優化效率
- 加入正則化–提高泛化效能
- 模組化設計
①generateds.py
# modelNN_generateds.py
# 資料匯入模組，生成模擬資料集
# coding: utf-8
import numpy as np
import matplotlib.pyplot as plt
import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2' #hide warnings
seed = 2
def generateds():
# 基於seed產生隨機數
rdm = np.random.RandomState(seed)
# 隨機數返回300行2列的矩陣，表示300組座標點，作為輸入資料集
X = rdm.randn(300, 2)
# 手工標註資料分類
Y_ = [int(x0*x0 + x1*x1 < 2)for (x0, x1) in X]
# Y_為1，標記紅色，否則藍色
Y_c = [['red' if y else 'blue'] for y in Y_]
# 對資料集和標籤進行reshape， X整理為n行2列，Y為n行1列，第一個元素-1表示n行
X = np.vstack(X).reshape(-1, 2)
Y_ = np.vstack(Y_).reshape(-1, 1)
return X, Y_, Y_c
print("X:\n")
print(X)
print("Y_:\n")
print(Y_)
print("Y_c:\n")
print(Y_c)
② forward.py
# modelNN_generateds.py
# 前向傳播模組
# 定義神經網路的輸入、引數和輸出，定義前向傳播過程
# coding: utf-8
import tensorflow as tf
import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2' #hide warnings
# 給w賦初值，並把w的正則化損失加到總損失中
def get_weight(shape, regularizer):
w = tf.Variable(tf.random_normal(shape), dtype = tf.float32)
tf.add_to_collection('losses', tf.contrib.layers.l2_regularizer(regularizer)(w))
return w
# 給b賦初值
def get_bias(shape):
b = tf.Variable(tf.constant(0.01, shape = shape))
return b
def forward(x, regularizer):
w1 = get_weight([2, 11], regularizer)
b1 = get_bias([11])
y1 = tf.nn.relu(tf.matmul(x, w1) + b1)
w2 = get_weight([11, 1], regularizer)
b2 = get_bias([1])
y = tf.matmul(y1, w2) + b2 #輸出層不通過啟用函式
return y
③ backward.py
# modelNN_generateds.py
# 反向傳播模組
# 定義神經網路的反向傳播過程
# coding: utf-8
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
import modelNN_generateds
import modelNN_forward
import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2' #hide warnings
# 定義超引數
STEPS = 40000 #訓練輪數
BATCH_SIZE = 30
LEARNING_RATE_BASE = 0.001 #初始學習率
LEARNING_RATE_DECAY = 0.999 # 學習率衰減率
REGULARIZER = 0.01 # 正則化引數
def backward():
# placeholder佔位
x = tf.placeholder(tf.float32, shape = (None, 2))
y_ = tf.placeholder(tf.float32, shape = (None, 1))
# 生成資料集
X, Y_, Y_c = modelNN_generateds.generateds()
# 前向傳播推測輸出y
y = modelNN_forward.forward(x, REGULARIZER)
# 定義global_step
global_step = tf.Variable(0, trainable = False)
# 定義指數衰減學習率
learning_rate = tf.train.exponential_decay(
LEARNING_RATE_BASE,
global_step,
300/BATCH_SIZE,
LEARNING_RATE_DECAY,
staircase = True)
# 定義損失函式
loss_mse = tf.reduce_mean(tf.square(y - y_))
loss_total = loss_mse + tf.add_n(tf.get_collection('losses'))
# 定義反向傳播方法：包含正則化
train_step = tf.train.AdamOptimizer(learning_rate).minimize(loss_total)
# 定義訓練過程
with tf.Session() as sess:
init_op = tf.global_variables_initializer()
sess.run(init_op)
for i in range(STEPS):
start = (i * BATCH_SIZE) % 300
end = start + BATCH_SIZE
sess.run(train_step, feed_dict = {x: X[start:end], y_:Y_[start:end]})
if i % 2000==0:
loss_v = sess.run(loss_total, feed_dict = {x: X, y_: Y_})
print("after %d steps, loss for total is %f" %(i, loss_v))
xx, yy = np.mgrid[-3:3:.01, -3:3:.01]
grid = np.c_[xx.ravel(), yy.ravel()]
probs = sess.run(y, feed_dict = {x: grid})
probs = probs.reshape(xx.shape)
# 視覺化
plt.scatter(X[:, 0], X[:, 1], c = np.squeeze(Y_c))
# 給probs值為0.5的所有點(xx, yy)上色
plt.contour(xx, yy, probs, levels = [.5])
plt.show()
# 判斷python執行檔案是否為主檔案，如果是，則執行
if __name__ == '__main__':
backward()
