PyTorch实现可视化一元线性回归的Demo

一、环境

OS：Ubuntu 18.04
Environment: PyTorch&Anaconda3
Editor: Spyder

二、代码部分

已更新神经网络代码----PyTorch实现可视化一元线性回归的Demo2

Spyder输入不了中文，请原谅我注释里的的Chinglish
一元线性回归的代码在网络上有很多，本代码demo做了如下更新：

添加了打印loss的switch
利用matplotlib实现了三个阶段的可视化图形和并可同时输出，方便对比
由交互式的Demo改成了可以解释到底的Demo，减少学习工作量
添加了我的进步汗水与失去的头发

学习不易，转载请注明出处。

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
@ubuntu 18.04
@spyder editor
@author: ftimes
"""
import torch
import numpy as np
from torch.autograd import Variable
import matplotlib.pyplot as plt

def getNpArray():
    return np.random.randn(15,1)

    
X = np.array([[3.3], [4.4], [5.5], [6.71], [6.93], [4.168],
                    [9.779], [6.182], [7.59], [2.167], [7.042],
                    [10.791], [5.313], [7.997], [3.1]], dtype=np.float32)

Y = np.array([[1.7], [2.76], [2.09], [3.19], [1.694], [1.573],
                    [3.366], [2.596], [2.53], [1.221], [2.827],
                    [3.465], [1.65], [2.904], [1.3]], dtype=np.float32) 

plt.title('initial value')
plt.figure(1)    
plt.plot(X, Y, 'bo')

#trans to tensor
X = torch.from_numpy(X)
Y = torch.from_numpy(Y)

#def pra a,b
a = Variable(torch.randn(1), requires_grad=True)
b = Variable(torch.zeros(1), requires_grad=True)

#trans to variable
X = Variable(X)
Y = Variable(Y)
def linear_model(x):
    return a*x + b

y_ = linear_model(X)
plt.figure(2)    
plt.title('append estiamted value')
plt.plot(X.data.numpy(), Y.data.numpy(), 'bo', label='real')
plt.plot(X.data.numpy(), y_.data.numpy(), 'ro', label='estimated')
plt.legend()

#compute and def loss
def computeloss(y_, Y):
    return torch.mean((y_ - Y) ** 2)

learningRate=1e-2#def rate of learning to contral model

#initialize
loss = computeloss(y_, Y)
loss.backward()
a.data=a.data-learningRate*a.grad.data
b.data=b.data-learningRate*b.grad.data
print('Need u print loss?    [y/n]')
order2=input()
lossflag=False
if order2=='y':
    lossflag=True    


times=100
for i in range(times):
    y_ = linear_model(X)
    loss = computeloss(y_, Y)
    
    a.grad.zero_() #must set zero 
    b.grad.zero_()
    loss.backward()
    #compute grad of a,b
    a.data=a.data-learningRate*a.grad.data
    b.data=b.data-learningRate*b.grad.data
    if lossflag:
         print('epoch: {}, loss: {}'.format(i, loss.data))

plt.figure(3)
plt.title('after update')
plt.plot(X.data.numpy(), Y.data.numpy(), 'bo', label='real')
plt.plot(X.data.numpy(), y_.data.numpy(), 'ro', label='estimated')
plt.legend()

三、示例

Need u print loss? [y/n]

y

epoch: 0, loss: 0.964177709961411
epoch: 1, loss: 0.9637709491045978
epoch: 2, loss: 0.9633792086085372
epoch: 3, loss: 0.963001928222742
epoch: 4, loss: 0.9626385689124767
epoch: 5, loss: 0.9622886116576249
epoch: 6, loss: 0.9619515570811104
epoch: 7, loss: 0.9616269249924927
epoch: 8, loss: 0.961314252842035
epoch: 9, loss: 0.9610130955541414
epoch: 10, loss: 0.9607230246097402
epoch: 11, loss: 0.9604436282531694
epoch: 12, loss: 0.9601745095184536
epoch: 13, loss: 0.9599152864984009
epoch: 14, loss: 0.9596655917972626
epoch:15, loss: 0.9594250718811523
epoch: 16, loss: 0.9591933860391304
epoch: 17, loss: 0.9589702067954244
epoch: 18, loss: 0.9587552187163422
epoch: 19, loss: 0.9585481182398691
epoch: 20, loss: 0.958348612958298