PyTorch遇到的error1

RuntimeError: Expected object of type torch.cuda.FloatTensor but found type torch.FloatTensor for argument #4 'mat1'

TypeError: can't convert CUDA tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.

源代码，廖星宇学习笔记

#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Fri Sep 14 08:22:04 2018

@author: lthpc
"""

import torch
from torch.autograd import Variable
from torch import nn,optim
import numpy as np
import matplotlib.pyplot as plt

x_train = 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_train = 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)
'''
#show
plt.scatter(x_train,y_train)
plt.show()
'''
'''
transform tensor
'''
x_train = torch.from_numpy(x_train)
y_train = torch.from_numpy(y_train)
#print('from numpy to torch.Tensor is {}'.format(x_train))
#x_train = x_train.float()
#y_train = y_train.float()
#print(x_train.type(torch.FloatTensor))

'''
defin model
'''
class LinearRegression(nn.Module):
    def __init__(self):
        super(LinearRegression,self).__init__()
        self.linear = nn.Linear(1,1)
        
    def forward(self,x):
        out = self.linear(x)
        return out

'''
GPU or not
'''
if torch.cuda.is_available():
    model = LinearRegression().cuda()
else:
    model = LinearRegression()
    
'''
optimizer
'''
criterion = nn.MSELoss()
optimizer = optim.SGD(model.parameters(),lr=1e-3)

'''
train
'''
num_epochs = 1000
for epoch in range(num_epochs):
    if torch.cuda.is_available():
        inputs = Variable(x_train).cuda()
        target = Variable(y_train).cuda()
    else:
        inputs = Variable(x_train)
        target = Variable(y_train)
#        forward
    out = model(inputs)
    loss = criterion(out,target)
#        backward
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()
    
    if (epoch+1) % 2==0:
        print('Epoch[{}/{}],loss:{:.6f}'
                     .format(epoch+1,num_epochs,loss.data[0]))
'''
predict
'''
model.eval()
#print(x_train.type())
predict = model(Variable(x_train))

predict = predict.data.numpy()
plt.plot(x_train.numpy(),y_train.numpy(),'ro',label='Original data')
plt.plot(x_train.numpy(),predict,label='Fitting Line')
plt.show()

原因1：训练时采用GPU的数据格式，将测试部分错误的地方改为predict = model(Variable(x_train).cuda())

原因2：如果想把CUDA tensor格式的数据改成numpy时，需要先将其转换成cpu float-tensor随后再转到numpy格式。改为predict = predict.data.cpu().numpy()

成功后显示拟合图片