TensorFlow大杂烩

TensorFlow是什么

TensorFlow是一个谷歌出品的深度学习框架

TensorFlow的安装

pip install tensorflow 大约需要下载100MB

TensorFlow之HelloWorld

示例: 使用TensorFlow在平面上计算最接近指定点集的直线

import tensorflow as tf

# 已知二维平面的五个点(训练数据)，计算与这五个点最接近的一条直线(最终待求模型)

points = [(1, 4), (2, 6), (3, 8), (5, 12), (8, 18)]

# 横轴数据、纵轴数据
xx = [point[0] for point in points]
yy = [point[1] for point in points]

# 直线公式: y = a * x + b

# 待求变量: 直线坡度a, 直线与y轴的交点b
a = tf.Variable(0, dtype=tf.float32)
b = tf.Variable(0, dtype=tf.float32)

# 样本占位符x, y
x = tf.placeholder(tf.float32)
y = tf.placeholder(tf.float32)

# 损失模型/误差模型
loss = tf.reduce_sum(tf.square(a * x + b - y))
# 训练模型(梯度0.001)
train = tf.train.GradientDescentOptimizer(0.001).minimize(loss)

# 创建TensorFlow会话
sess = tf.Session()
# 初始化TensorFlow全局变量
sess.run(tf.global_variables_initializer())
# 训练5000次
for i in range(5000):
    sess.run(train, {x: xx, y: yy})
    if (i < 10) or (i % 1000 is 0):
        print(f"{i + 1:04}: y={sess.run(a):.9f}x+{sess.run(b):.9f} => {sess.run(loss, {x: xx, y: yy}):.12f}")

控制台输出

输出格式: 训练次数: 公式 => 损失

0001: y=0.488000035x+0.096000001 => 362.995117187500
0002: y=0.871824026x+0.172496006 => 226.141662597656
0003: y=1.173673511x+0.233641744 => 141.394470214844
0004: y=1.411018372x+0.282705724 => 88.911415100098
0005: y=1.597605824x+0.322259963 => 56.406211853027
0006: y=1.744253159x+0.354328334 => 36.271305084229
0007: y=1.859472513x+0.380503446 => 23.796068191528
0008: y=1.949962020x+0.402038455 => 16.063734054565
0009: y=2.020992279x+0.419919521 => 11.268218040466
0010: y=2.076710939x+0.434922606 => 8.291220664978
1001: y=2.016209364x+1.913361430 => 0.011228192598
2001: y=2.000897169x+1.995205998 => 0.000034384284
3001: y=2.000049353x+1.999734759 => 0.000000105264
4001: y=2.000005484x+1.999973178 => 0.000000001114

HelloWorld之NumPy版本

import tensorflow as tf
import numpy as np

# 以公式(y=2x+2)构造包含10个元素的点集
points = np.array([(x, 2 * x + 2) for x in range(11) if x is not 0])
print(f"Points: \n{points}")

xData = points[:, 0]
yData = points[:, 1]
print(f"X-Axis: {xData}\nY-Axis: {yData}")

# 直线的斜截式方程: y = Ax + B
# Ax + B - y 越接近0，直线越拟合点集
A = tf.Variable(0.0)  # 斜率
B = tf.Variable(0.0)  # 截距

train = tf.train.GradientDescentOptimizer(0.001).minimize(tf.reduce_sum(tf.square(A * xData + B - yData)))

sess = tf.Session()
sess.run(tf.global_variables_initializer())
for _ in range(1000):
    if _ % 100 is 0:
        print(f"{_:03} A = {sess.run(A):.9f}, B = {sess.run(B):.9f}")
    sess.run(train)

控制台输出:

Points:
[[ 1  4]
 [ 2  6]
 [ 3  8]
 [ 4 10]
 [ 5 12]
 [ 6 14]
 [ 7 16]
 [ 8 18]
 [ 9 20]
 [10 22]]
X-Axis: [ 1  2  3  4  5  6  7  8  9 10]
Y-Axis: [ 4  6  8 10 12 14 16 18 20 22]
000 A = 0.000000000, B = 0.000000000
100 A = 2.158243656, B = 0.898336709
200 A = 2.103886366, B = 1.276762247
300 A = 2.068200827, B = 1.525196910
400 A = 2.044773579, B = 1.688293815
500 A = 2.029393673, B = 1.795366049
600 A = 2.019296646, B = 1.865659118
700 A = 2.012668371, B = 1.911805630
800 A = 2.008316755, B = 1.942100525
900 A = 2.005459785, B = 1.961989641

文章首发: https://baijifeilong.github.io