win7 安装PyTorch

win7 安装PyTorch并测试

4月25号，PyTorch官方支持windows安装，博主第一时间安装使用了，下面分享安装过程和测试案例。

1. 安装anaconda，方法查看：https://blog.csdn.net/zyb228/article/details/77995520
   版本选择：anaconda3，python3.5的就可以

2. 安装PyTorch:
   

由于conda安装很慢，而且出错，所以本文采用pip3安装，如下图：

测试：

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


# 超参数
input_size = 1
output_size = 1
num = 60
learning_rate = 0.002

# 构造数据
x_train = np.array([ [6.93], [4.168], [4.2], [5.5],[3.3], [6.71], 
                    [9.79], [6.3], [7.2],  [5.13], [7.997],
                    [10.71],  [7.59], [2.167],[3.1]], dtype=np.float32)

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

# 线性回归模型(Linear Regression Model)
class LRegression(nn.Module):
    def __init__(self, input_size, output_size):
        super(LRegression, self).__init__()
        self.linear = nn.Linear(input_size, output_size)  
    
    def forward(self, x):
        out = self.linear(x)
        return out

model = LRegression(input_size, output_size)

loss_criterion = nn.MSELoss()
#优化器Optimizer
optimizer = torch.optim.SGD(model.parameters(), lr=learning_rate)  

# 训练模型
for i in range(num):
    # 将ndarray转换为Variabl
    inputs = Variable(torch.from_numpy(x_train))
    targets = Variable(torch.from_numpy(y_train))

    #前向传播
    optimizer.zero_grad()  
    outputs = model(inputs)
    loss = loss_criterion(outputs, targets)
    #反向传播
    loss.backward()
    #优化
    optimizer.step()
    
    if (i+1) % 5 == 0:
        print ('%d/%d, Loss: %.4f' 
               %(i+1, num, loss.data[0]))
        
# 显示
inputs=Variable(torch.from_numpy(x_train))
pred = model(inputs).data.numpy()
plt.plot(x_train, y_train, 'ro', label='Original data')
plt.plot(x_train, pred, label='Fitted line')
plt.legend()
plt.show()

注：代码来源于网络
效果：

最新版本的Pytorch 1.3 安装测试案例，
训练模块代码由 博主 lyjiang1 根据最新安装的测试情况进行更新。
感谢博主 lyjiang1！

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


# 超参数
input_size = 1
output_size = 1
num = 60
learning_rate = 0.002

# 构造数据
x_train = np.array([ [6.93], [4.168], [4.2], [5.5],[3.3], [6.71], 
                    [9.79], [6.3], [7.2],  [5.13], [7.997],
                    [10.71],  [7.59], [2.167],[3.1]], dtype=np.float32)

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

# 线性回归模型(Linear Regression Model)
class LRegression(nn.Module):
    def __init__(self, input_size, output_size):
        super(LRegression, self).__init__()
        self.linear = nn.Linear(input_size, output_size)  
    
    def forward(self, x):
        out = self.linear(x)
        return out

model = LRegression(input_size, output_size)

loss_criterion = nn.MSELoss()
#优化器Optimizer
optimizer = torch.optim.SGD(model.parameters(), lr=learning_rate)  

# 训练模型
#这段训练模型代码 是由 lyjiang1 根据最新版本的Pytorch 修改的测试案例
# 再次特别感谢 lyjiang1 ！   
for i in range(num):
    # 将ndarray转换为Variabl
    inputs = Variable(torch.from_numpy(x_train))
    targets = Variable(torch.from_numpy(y_train))
 
    #前向传播
    optimizer.zero_grad()  
    outputs = model(inputs)
    loss = loss_criterion(outputs, targets)
    #反向传播
    loss.backward()
    #优化
    optimizer.step()
    #a=loss.data[0] 老版本
    a=loss.item() #新版本
    if (i+1) % 5 == 0:
        print ('%d/%d, Loss: %.4f' % (i+1, num, a))

# 显示
inputs=Variable(torch.from_numpy(x_train))
pred = model(inputs).data.numpy()
plt.plot(x_train, y_train, 'ro', label='Original data')
plt.plot(x_train, pred, label='Fitted line')
plt.legend()
plt.show()

更多TensorFlow和PyTorch技术干货请关注：