1. Johngo学长首页
  2. 技术点详细复盘专题
  3. 人工智能

【一些笔记】TensorFlow笔记

• 人工智能 • 阅读 76

TensorFlow笔记

[by_041]

TensorFlow是基于 Tensor（张量）计算的一种 深度学习库
参考B站视频
一个博主，他最开始的博文全是关于TF的（至少22篇）
Tensorflow 1 的官方文档 用py2的

文章目录

安装

  • 使用 Anaconda Prompt
  • 先升级pip
    python -m pip install --upgrade pip
  • 再用镜像站安装
    pip install tensorflow-cpu -i https://pypi.douban.com/simple/ 不知道啥玩意的（跳过）
    pip install numpy pandas matplotlib sklearn tensorflow==2.0.0-alpha0 -i http://pypi.douban.com/simple/ 安装notebook配合使用体验更佳（体现在Jupyter中的使用）
    pip install notebook -i http://pypi.douban.com/simple/
  • 启用
    activate tensorflow 禁用
    deactivate
  • 快速安装
（若需要升级pip）python -m pip install --upgrade pip
pip install tensorflow-cpu -i https://pypi.douban.com/simple/
pip install notebook -i http://pypi.douban.com/simple/
activate tensorflow
  • 尝试以下程序。如果你能运行它，它就可以了。
    [En]

    try the following program. If you can run it, it will be OK.*

import osos.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'import tensorflow as tft0 = tf.constant(3, dtype=tf.int32)t1 = tf.constant([3., 4.1, 5.2], dtype=tf.float32)t2 = tf.constant([['Apple', 'Orange'], ['Potato', 'Tomato']], dtype=tf.string)t3 = tf.constant([[[5], [6], [7]], [[4], [3], [2]]])print('==========================================================================')print(t0)print('==========================================================================')print(t1)print('==========================================================================')print(t2)print('==========================================================================')print(t3)print('==========================================================================')

语法基础

TensorFlow是基于 Tensor（张量）计算的一种深度学习库，其基础语法都有所体现

使用基础

数据处理基础

包括用 pandas读取数据

基于 pandas 进行读取数据

import pandas as pd
data = pd.read_csv('C:\\Users\\15614\\Desktop\\应用基础\\数模实战\\数据\\一手粉丝量 - 90 days.csv')
print(data)

基于 matplotlib 进行数据绘图


import matplotlib.pyplot as plt

plt.scatter(data.time,data.fans)

Tensorflow 2.0 tf.keras

官方推荐的首选

一次完整的线性回归分析


import pandas as pd
data = pd.read_csv('C:\\Users\\15614\\Desktop\\应用基础\\数模实战\\数据\\一手粉丝量 - 90 days.csv')
print(data)

import matplotlib.pyplot as plt

plt.scatter(data.time,data.fans)

x=data.time
y=data.fans

import tensorflow as tf

model=tf.keras.Sequential()
model.add(tf.keras.layers.Dense(1,input_shape=(1,)))

print('==========================================================================')
print(model.summary())
print('==========================================================================')

model.compile(optimizer='adam',
    loss='mse'
    )

history=model.fit(x,y,epochs=500)

predict_1=model.predict(x)

predict_2=model.predict(pd.Series([91,92,93]))

print('现有预测：')
print(predict_1)
print('预测未来3天：')
print(predict_2)

简化后


import pandas as pd
data = pd.read_csv('C:\\Users\\15614\\Desktop\\应用基础\\数模实战\\数据\\一手粉丝量 - 90 days.csv')
print(data)

import matplotlib.pyplot as plt

plt.scatter(data.time,data.fans)
plt.show()

x=data.time
y=data.fans

import tensorflow as tf

model=tf.keras.Sequential()
model.add(tf.keras.layers.Dense(1,input_shape=(1,)))

model.compile(optimizer='adam',
    loss='mse'
    )

history=model.fit(x,y,epochs=5000)

model.predict(pd.Series([91,92,93]))
  • 自己装好了，跳过。

My改进版

  • 预测模型：y = w × x + b y=w\times x+ b y =w ×x +b
import random as rd
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

eps=1e-6

np.random.seed(123)
x=np.random.random(100)*100

y=np.random.random(100)*340-666
x.sort()
y.sort()

plt.scatter(x, y)

w, b = 0, 0
num_epoch = 10000
learning_rate = 0.000005
loss=0
e=0
cnt_pass=0
while True :

    e+=1
    las_loss=loss
    las_w=w
    las_b=b

    y_pred = w * x + b

    loss = np.mean(np.square(y_pred-y))

    grad_w, grad_b = (y_pred - y).dot(x), (y_pred - y).sum()

    w, b = w - learning_rate * grad_w, b - learning_rate * grad_b

    if abs(las_loss-loss)<eps and abs(las_w-w)<eps and abs(las_b-b)<eps :
        cnt_pass+=1
        if cnt_pass>100:
            break
    else:
        cnt_pass=0
    if e %2000 ==0:
        plt.cla()
        plt.scatter(x,y)
        plt.plot(x, y_pred, 'r-', lw=5)
        plt.title('Loss=%.4f | adj=%e | w=%.4f | b=%.4f'%(loss,learning_rate,w,b),
            fontdict={'size': 14, 'color':  'red'})
        plt.pause(0.1)
plt.cla()
plt.scatter(x,y)
plt.plot(x, y_pred, 'r-', lw=5)
plt.title('Done < at e=%d >\nLoss=%.4f | adj=%e | w=%.4f | b=%.4f'%(e,loss,learning_rate,w,b),
    fontdict={'size': 14, 'color':  'red'})
plt.pause(0.1)
plt.show()
print(w, b)

乱搞的二次函数

  • 预测模型：y = w × x 2 + b y=w\times x^2+ b y =w ×x 2 +b
import random as rd
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

eps=1e-6

np.random.seed(123)

x=np.random.random(100)*100-50
x.sort()
y=np.array([x[i]**2+rd.random()*180. for i in range(100)])

plt.scatter(x, y)

xx=np.array([(x[i])**2*1. for i in range(100)])

w, b = 0, 0
num_epoch = 10000
learning_rate = 0.00000005
for e in range(num_epoch or True):

    y_pred =w*xx+b

    loss = np.mean(np.square(y_pred-y))

    grad_w, grad_b = (y_pred - y).dot(x), (y_pred - y).sum()

    las_w=w
    las_b=b
    w, b = w - learning_rate * grad_w, b - learning_rate * grad_b
    if abs(las_w-w)<eps and abs(las_b-b)<eps :
        break
    if e %70 ==0:
        plt.cla()
        plt.scatter(x,y)
        plt.plot(x, y_pred, 'r-', lw=5)
        plt.title('Loss=%.4f | adj=%e | w=%.4f | b=%.4f'%(loss,learning_rate,w,b),
            fontdict={'size': 14, 'color':  'red'})
        plt.pause(0.1)
plt.cla()
plt.scatter(x,y)
plt.plot(x, y_pred, 'r-', lw=5)
plt.title('Done\nLoss=%.4f | adj=%e | w=%.4f | b=%.4f'%(loss,learning_rate,w,b),
    fontdict={'size': 14, 'color':  'red'})
plt.pause(0.1)
plt.show()
print(w, b)
  • 预测模型：y = w × ( x − 50 ) 2 + b y=w\times (x-50)^2+ b y =w ×(x −5 0 )2 +b
import random as rd
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

eps=1e-6

np.random.seed(123)

x=np.random.random(100)*100
x.sort()
y=np.array([(x[i]-50)**2+rd.random()*180. for i in range(100)])

plt.scatter(x, y)

xx=np.array([(x[i]-50)**2*1. for i in range(100)])

w, b = 0, 0
num_epoch = 10000
learning_rate = 0.0000000005
for e in range(num_epoch or True):

    y_pred =w*xx+b

    loss = np.mean(np.square(y_pred-y))

    grad_w, grad_b = (y_pred - y).dot(x), (y_pred - y).sum()

    las_w=w
    las_b=b
    w, b = w - learning_rate * grad_w, b - learning_rate * grad_b
    if abs(las_w-w)<eps and abs(las_b-b)<eps :
        break
    if e %70 ==0:
        plt.cla()
        plt.scatter(x,y)
        plt.plot(x, y_pred, 'r-', lw=5)
        plt.title('Loss=%.4f | adj=%e | w=%.4f | b=%.4f'%(loss,learning_rate,w,b),
            fontdict={'size': 14, 'color':  'red'})
        plt.pause(0.1)
plt.cla()
plt.scatter(x,y)
plt.plot(x, y_pred, 'r-', lw=5)
plt.title('Done\nLoss=%.4f | adj=%e | w=%.4f | b=%.4f'%(loss,learning_rate,w,b),
    fontdict={'size': 14, 'color':  'red'})
plt.pause(0.1)
plt.show()
print(w, b)

Original: https://blog.csdn.net/qq_42710619/article/details/124246996
Author: 青菜 – Teloy_041
Title: 【一些笔记】TensorFlow笔记

原创文章受到原创版权保护。转载请注明出处：https://www.johngo689.com/508823/

转载文章受原作者版权保护。转载请注明原作者出处！

赞 (0)
0

【自取】最近整理的，有需要可以领取学习：



Linux核心资料大放送~

全栈面试题汇总（持续更新&可下载）

一个提高学习100%效率的工具！

【超详细】深度学习面试题目！

LeetCode Python刷题答案下载！

LeetCode Java版刷题答案下载！

LeetCode C++ 版本，抓紧保存！

LeetCode GO语言 刷题答案下载！

大家都在看

亲爱的 Coder【最近整理，可免费获取】👉 最新必读书单  | 👏 面试题下载  | 🌎 免费的AI知识星球