PyTorch的学习笔记
PyTorch的Tensor
官网上是类比Numpy的Ndarrays，相比于Ndarrays多了GPU相关的加速运算。关于Ndarrays其实就是N-dimension Arrays，是一个N维的数组对象。Pytorch里提供的许多方法也与Numpy很像。
下面提供了关于创建矩阵的方法，包括empty，rand，zeros，tensor等。
torch创建的对象是tensor。
torch.empty()
empty方法创建的矩阵不是空矩阵，而是未初始化的矩阵，所以里面的值不一定是0。
import torch
x = torch.empty(5,3)
print(x)
tensor([[ 1.1018e-08, 4.5818e-41, -1.2501e+11],
[ 4.5916e-41, 4.9724e-17, 4.5818e-41],
[ 4.9696e-17, 4.5818e-41, 1.3498e-08],
[ 4.5818e-41, 1.3416e-08, 4.5818e-41],
[ 4.9691e-17, 4.5818e-41, 4.9691e-17]])
torch.rand()
rand方法生成的是一个用随机数初始化的矩阵，里面的数值为限定时是在[0,1]之间随机生成
x = torch.rand(3,5)
print(x)
tensor([[ 0.9101, 0.5218, 0.6148, 0.2900, 0.3983],
[ 0.8021, 0.3228, 0.2530, 0.3052, 0.6225],
[ 0.0999, 0.2756, 0.8993, 0.1512, 0.0917]])
torch.zeros()
zeros方法顾名思义是生成0矩阵，detype参数指定了生成的数据的类型
x = torch.zeros(5, 3, dtype=torch.long)
print(x)
tensor([[ 0, 0, 0],
[ 0, 0, 0],
[ 0, 0, 0],
[ 0, 0, 0],
[ 0, 0, 0]])
torch.tensor()
tensor 方法能直接生成tensor构成的数据（这里有点迷），不过生成的数据自动保留四位小数
x = torch.tensor([5.5, 3,9.99999,8.16])
print(x)
tensor([ 5.5000, 3.0000, 10.0000, 8.1600])
从现成的tensor来生成新的tensor
x = x.new_ones(5, 3, dtype=torch.double) # new_* methods take in sizes
print(x)
x = torch.randn_like(x, dtype=torch.float) # override dtype!
print(x) # result has the same size
tensor([[ 1., 1., 1.],
[ 1., 1., 1.],
[ 1., 1., 1.],
[ 1., 1., 1.],
[ 1., 1., 1.]], dtype=torch.float64)
tensor([[ 0.2710, 1.6084, -0.8171],
[ 0.8977, 1.4150, 0.2287],
[ 0.2518, -0.0245, -0.5036],
[-0.5529, -0.1147, -1.3930],
[-0.7114, 0.2698, 2.3081]])
Torch里面重载的一些基本运算操作
矩阵的+法运算
加法运算要求矩阵的大小相同，dtype也需要相同才能进行运算,add方法与所重载的+号类似。
add方法的接受参数
- (Tensor input, Tensor other, float alpha, Tensor out)
- (Tensor input, float other, float alpha, Tensor out)
y = torch.rand(5,3,dtype=torch.float)
print(y)
z = x + y
print(z)
tensor([[ 0.1392, 0.6817, 0.1232],
[ 0.6517, 0.5228, 0.1931],
[ 0.8584, 0.7995, 0.4617],
[ 0.1270, 0.0686, 0.8771],
[ 0.1968, 0.9849, 0.8087]])
tensor([[ 0.4103, 2.2900, -0.6938],
[ 1.5495, 1.9379, 0.4218],
[ 1.1102, 0.7750, -0.0419],
[-0.4260, -0.0461, -0.5159],
[-0.5146, 1.2546, 3.1168]])
z = torch.add(x , y)
print(z)
tensor([[ 0.4103, 2.2900, -0.6938],
[ 1.5495, 1.9379, 0.4218],
[ 1.1102, 0.7750, -0.0419],
[-0.4260, -0.0461, -0.5159],
[-0.5146, 1.2546, 3.1168]])
output = torch.empty(5,3)
print(output)
torch.add(x,y,out=output)
print(output)
tensor(1.00000e-08 *
[[ 1.1018, 0.0000, 1.1018],
[ 0.0000, 0.0000, 0.0000],
[ 0.0000, 0.0000, 0.0000],
[ 0.0000, 0.0000, 0.0000],
[ 0.0000, 0.0000, 0.0000]])
tensor([[ 0.4103, 2.2900, -0.6938],
[ 1.5495, 1.9379, 0.4218],
[ 1.1102, 0.7750, -0.0419],
[-0.4260, -0.0461, -0.5159],
[-0.5146, 1.2546, 3.1168]])
替换性操作
在提供的方法中，如果是(operation)_ (arg)格式的方法的，操作后会替换到调用这个方法的对象
y.add_(x)
print(y)
tensor([[ 0.4103, 2.2900, -0.6938],
[ 1.5495, 1.9379, 0.4218],
[ 1.1102, 0.7750, -0.0419],
[-0.4260, -0.0461, -0.5159],
[-0.5146, 1.2546, 3.1168]])
x.copy_(y)
print(x)
tensor([[ 0.4103, 2.2900, -0.6938],
[ 1.5495, 1.9379, 0.4218],
[ 1.1102, 0.7750, -0.0419],
[-0.4260, -0.0461, -0.5159],
[-0.5146, 1.2546, 3.1168]])
类Numpy的索引方式
print(x[:,1])
tensor([ 2.2900, 1.9379, 0.7750, -0.0461, 1.2546])
torch.view()
改变矩阵的大小，类似resize，reshape
x = torch.randn(4, 4)
y = x.view(16)
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])
print(x,y,z)
tensor([[-0.3858, -0.6874, 1.0538, -1.2053],
[-0.2992, 0.7963, 0.5221, -1.0758],
[-1.1194, 0.1516, 2.0523, 0.4788],
[ 0.3233, -0.0533, 0.3937, -2.0091]]) tensor([-0.3858, -0.6874, 1.0538, -1.2053, -0.2992, 0.7963, 0.5221,
-1.0758, -1.1194, 0.1516, 2.0523, 0.4788, 0.3233, -0.0533,
0.3937, -2.0091]) tensor([[-0.3858, -0.6874, 1.0538, -1.2053, -0.2992, 0.7963, 0.5221,
-1.0758],
[-1.1194, 0.1516, 2.0523, 0.4788, 0.3233, -0.0533, 0.3937,
-2.0091]])
对于单值的tensor可用.item()方法获取其详细的值。
x = torch.rand(1)
print(x)
print(x.item())
tensor([ 0.5580])
0.5579903721809387
从Torch的Tensor到Numpy的转换
numpy方法
x = torch.ones(5)
print(x)
y = x.numpy()
print(y)
tensor([ 1., 1., 1., 1., 1.])
[1. 1. 1. 1. 1.]
x.add_(1)
print(x)
print(y)
tensor([ 2., 2., 2., 2., 2.])
[2. 2. 2. 2. 2.]
可以发现对x操作会影响到y
从Numpy到Tensor的转换
import numpy as np
a = np.ones(5)
b = torch.from_numpy(a)
np.add(a, 1, out=a)
print(a)
print(b)
[2. 2. 2. 2. 2.]
tensor([ 2., 2., 2., 2., 2.], dtype=torch.float64)
可以看到从numpy生成的tensor也会共同影响
Reference:
