问回归不起作用的简单案例(PyTorch)

EN

####
#code from https://medium.com/@benjamin.phillips22/simple-regression-with-neural-networks-in-pytorch-313f06910379
####

import torch
from torch.autograd import Variable
import torch.nn.functional as F
import torch.utils.data as Data
import matplotlib.pyplot as plt
# %matplotlib inline
import numpy as np
import imageio
import pandas as pd

torch.manual_seed(1)    # reproducible

x=torch.from_numpy(pd.read_csv('./x.csv').to_numpy())
y=torch.from_numpy(pd.read_csv('./y.csv').to_numpy())
x=x.float()
y=y.float()

# torch can only train on Variable, so convert them to Variable
x, y = Variable(x), Variable(y)

# view data
plt.figure(figsize=(10,4))
plt.scatter(x.data.numpy(), y.data.numpy(), color = "orange")
plt.title('regression')
plt.xlabel('x input')
plt.ylabel('y output')
plt.show()


net = torch.nn.Sequential(
 torch.nn.Linear(1,20),
 torch.nn.Sigmoid(),
 torch.nn.Linear(20,20),
 torch.nn.Sigmoid(),
 torch.nn.Linear(20,1),
)
optimizer = torch.optim.SGD(net.parameters(), lr=0.1)
loss_func = torch.nn.MSELoss()  # this is for regression mean squared loss

my_images = []
fig, ax = plt.subplots(figsize=(12,7))

# train the network
for t in range(500):
  
    prediction = net(x)     # input x and predict based on x

    loss = loss_func(prediction, y)     # must be (1. nn output, 2. target)

    optimizer.zero_grad()   # clear gradients for next train
    loss.backward()         # backpropagation, compute gradients
    optimizer.step()        # apply gradients
    
    # plot and show learning process
    plt.cla()
    ax.set_xlabel('x input',fontsize=12)
    ax.set_ylabel('y output',fontsize=12)
    ax.set_xlim(min(x),max(x))
    ax.set_ylim(min(y),max(y))
    ax.scatter(x.data.numpy(), y.data.numpy(), color = "orange")
    ax.plot(x.data.numpy(), prediction.data.numpy(), 'g-', lw=3)
    ax.set_title('regression step='+str(t)+" loss="+str(loss.data.numpy()),fontsize=16)

    # Used to return the plot as an image array 
    # (https://ndres.me/post/matplotlib-animated-gifs-easily/)
    fig.canvas.draw()       # draw the canvas, cache the renderer
    image = np.frombuffer(fig.canvas.tostring_rgb(), dtype='uint8')
    image  = image.reshape(fig.canvas.get_width_height()[::-1] + (3,))

    my_images.append(image)
    print(str(t)+" "+str(loss.data.numpy()))
   
# save images as a gif    
imageio.mimsave('./result.gif', my_images, fps=10)

0.316905
0.322015
0.332582
0.310465
0.250653
0.292722
0.297352
0.276525
0.308451
0.283465
0.276011
0.29934
0.307176
0.28573
0.248667
0.288614
0.249795
0.255556
0.258393
0.235972
0.225121
0.207828
0.190252
0.192881
0.204084
0.167646
0.155202
0.146516
0.162462
0.182906
0.160287
0.213769
0.186362
0.201151
0.186125
0.190625
0.146851
0.169204
0.207855
0.196557
0.208835
0.2244
0.206303
0.193485
0.185266
0.205616
0.229315
0.196254
0.219849
0.209988
0.197361
0.195402
0.210149
0.240754
0.210418
0.191776
0.189532
0.206153
0.165696
0.187938
0.157561
0.163148
0.19473
0.18966
0.162334
0.189277
0.166506
0.198193
0.157867
0.135192
0.152216
0.137521
0.142007
0.121252
0.136517
0.118812
0.126124
0.141713
0.13222
0.2032
0.156077
0.166526
0.167117
0.130817
0.167058
0.188566
0.178803
0.224779
0.217089
0.194542
0.199796
0.246194
0.249908
0.23034
0.204611
0.222958
0.24259
0.234767
0.278205
0.267297
0.275127
0.264059
0.25439
0.287421
0.267725
0.252964
0.256326
0.229031
0.276914
0.244985
0.273892
0.298103
0.256733
0.27219
0.301747
0.278291
0.274979
0.300091
0.310184
0.333836
0.297877
0.279405
0.278263
0.291442
0.278518
0.28268
0.321826
0.355584
0.315503
0.338342
0.39687
0.388692
0.353228
0.368169
0.328025
0.407137
0.38092
0.357814
0.362786
0.405149
0.354694
0.348222
0.295455
0.307671
0.290612
0.24626
0.229377
0.26535
0.217139
0.206268
0.230013
0.255796
0.27014
0.246626
0.224845
0.272181
0.201281
0.252555
0.270198
0.289443
0.243552
0.238465
0.207842
0.197373
0.238857
0.224703
0.259659
0.288809
0.24757
0.264744
0.250775
0.245659
0.193861
0.296178
0.234242
0.219704
0.264879
0.290614
0.296195
0.237291
0.208546
0.295197
0.272628
0.288054
0.293539
0.305374
0.328142
0.328574
0.307407
0.298107
0.315636
0.301924