PyTorch – 安装和入门

Contents

1 安装PyTorch
2 测试GPU是否可用
3 PyTorch使用指定GPU
4 张量的使用
5 参考资料

安装PyTorch

官网的安装方式 https://pytorch.org/get-started/locally/
从pypi.org安装(GPU版本)
https://pypi.org/project/torch/#files
https://pypi.org/project/torchvision/#files
选择自己的python和cuda版本，比如python3.5就选择torch-1.2.0-cp35-cp35m-manylinux1_x86_64.whl
cuda9.0选择兼容版本torch==0.4.1
使用 torch.__version__ 查看版本号

测试GPU是否可用

import torch
flag = torch.cuda.is_available()
print(flag)
# True

device = torch.device("cuda:0" if flag else "cpu")
print(torch.cuda.get_device_name(0))
# GeForce GTX 1060 6GB
print(torch.rand(3,3).cuda()) 
'''
tensor([[0.0284, 0.8647, 0.0298],
    [0.3623, 0.6082, 0.9034],
    [0.5146, 0.1420, 0.0223]], device='cuda:0')
'''

import torch

flag = torch.cuda.is_available()

print(flag)

# True

device = torch.device("cuda:0" if flag else "cpu")

print(torch.cuda.get_device_name(0))

# GeForce GTX 1060 6GB

print(torch.rand(3,3).cuda())

'''

tensor([[0.0284, 0.8647, 0.0298],

[0.3623, 0.6082, 0.9034],

[0.5146, 0.1420, 0.0223]], device='cuda:0')

'''

PyTorch使用指定GPU

1.在终端中指定

CUDA_VISIBLE_DEVICES=1 python my_script.py

1 2	CUDA_VISIBLE_DEVICES=1 python my_script.py

2.使用Python代码

import os
os.environ["CUDA_VISIBLE_DEVICES"] = "2"

import os

os.environ["CUDA_VISIBLE_DEVICES"] = "2"

3.使用函数set_device

import torch
gpu_id = 0
torch.cuda.set_device(gpu_id)

import torch

gpu_id = 0

torch.cuda.set_device(gpu_id)

张量的使用

x = torch.tensor([5.5, 3])  # 常量
y = torch.rand(5, 3)  # 随机数
print(x.size())  # 大小

x = torch.randn(4, 4)
y = x.view(16)  # 改变大小
z = x.view(-1, 8)  # the size -1 is inferred from other dimensions
print(x.size(), y.size(), z.size())
'''
torch.Size([4, 4]) torch.Size([16]) torch.Size([2, 8])
'''

a = torch.ones(5)
b = a.numpy()  # tensor转换成numpy的类型
print(b)

c = torch.from_numpy(b)  # numpy类型转tensor
np.add(c, 1, out=c)

# 转成gpu变量
c.to('cuda:0')
c.cuda()

# 转成cpu变量
c.to('cpu')

x = torch.tensor([5.5, 3]) # 常量

y = torch.rand(5, 3) # 随机数

print(x.size()) # 大小

x = torch.randn(4, 4)

y = x.view(16) # 改变大小

z = x.view(-1, 8) # the size -1 is inferred from other dimensions

print(x.size(), y.size(), z.size())

'''

torch.Size([4, 4]) torch.Size([16]) torch.Size([2, 8])

'''

a = torch.ones(5)

b = a.numpy() # tensor转换成numpy的类型

print(b)

c = torch.from_numpy(b) # numpy类型转tensor

np.add(c, 1, out=c)

# 转成gpu变量

c.to('cuda:0')

c.cuda()

# 转成cpu变量

c.to('cpu')

参考资料

官方网站 https://pytorch.org/
中文文档 https://pytorch-cn.readthedocs.io/zh/latest/
官网教程中文版 http://pytorch123.com/
kaggle教程 https://www.kaggle.com/kanncaa1/pytorch-tutorial-for-deep-learning-lovers

X.YU