can‘t convert cuda:0 device type tensor to numpy. Use Tensor. cpu() to copy the tensor to host memory

Source code ：

#  Training curve 
result1 = [100 - i['train'] for i in statistics]
result2 = [100 - i['valid'] for i in statistics]
plt.figure(figsize = (10, 7))
plt.plot(result1, label = 'Training')
plt.plot(result2, label = 'Validation')
plt.xlabel('Step')
plt.ylabel('Error Rate')
plt.legend()

Error message ：

 

The wrong content is unable to draw normally , This error message is the same as the one above .cpu() In the same , But the solution is slightly different , I found many ways on the Internet , But no suitable solution has been found , Most of the Internet is talking about adding .cpu(), But as far as this drawing program is concerned , Unable to find a suitable addition .cpu() The place of . The reason is , Still don't really understand the meaning of the program .

First, print the traversed statistics, Look at what's in it .

Pictured above , You can find statistic Itself is a list , And the elements inside are indeed dictionaries , In the dictionary train,valid The corresponding value yes torch Of tensor, That's why there are mistakes .

Change it as follows ：

#  Training curve 
result1 = [100 - i['train'].cpu() for i in statistics]
result2 = [100 - i['valid'].cpu() for i in statistics]
plt.figure(figsize = (10, 7))
plt.plot(result1, label = 'Training')
plt.plot(result2, label = 'Validation')
plt.xlabel('Step')
plt.ylabel('Error Rate')
plt.legend()

  stay i['train'] and i['valid'] Add .cpu(), Click run again , You can run the program normally .