【Pytorch】Torch_Basic_Learning

前言

本篇为在.ipynb页面上的自学尝试记录,
可以在本人的个人主页上查看或下载自行测试,关于Windows10如何配置Pytorch请移步前篇文章即可~

  • 传送到 个人主页 (Github Pages)
  • 传送到 个人主页 (国内镜像源,加载速度稍快)
  • 传送到 Ipynb的查看或下载

Pytorch

  • Tensor computation (like numpy) with strong GPU acceleration
  • PyTorch is an optimized tensor library for deep learning using GPUs and CPUs.
  • It has a CUDA counterpart, that enables you to run your tensor computations on an NVIDIA GPU with compute capability >= 3.0.
1
2
3
4
5
6
7
8
9
10
from __future__ import print_function
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torch.autograd as autograd
from torch.autograd import Variable

import numpy as np
print("Import success.\nTorch Version:{}".format(torch.__version__))
Import success.
Torch Version:0.2.1+a4fc05a
Package Description
torch a Tensor library like NumPy, with strong GPU support
torch.autograd a tape based automatic differentiation library that supports all differentiable Tensor operations in torch
torch.nn a neural networks library deeply integrated with autograd designed for maximum flexibility
torch.optim an optimization package to be used with torch.nn with standard optimization methods such as SGD, RMSProp, LBFGS, Adam etc.
torch.multiprocessing python multiprocessing, but with magical memory sharing of torch Tensors across processes. Useful for data loading and hogwild training.
torch.utils DataLoader, Trainer and other utility functions for convenience
torch.legacy(.nn/.optim) legacy code that has been ported over from torch for backward compatibility reasons

Brief view in pytorch

1
2
3
4
5
6
7
8
torch.set_printoptions(
precision=None, # Number of digits of precision for floating point output (default 8).
threshold=None, # Total number of array elements which trigger summarization rather than full repr (default 1000).
edgeitems=None, # Number of array items in summary at beginning and end of each dimension (default 3).
linewidth=None, # The number of characters per line for the purpose of inserting line breaks (default 80).
# Thresholded matricies will ignore this parameter.
profile=None, # Sane defaults for pretty printing. Can override with any of the above options. (default, short, full)
)
1
2
3
4
5
6
7
8
9
np_mat = np.random.randn(1,2,3,4)
tc_mat = torch.randn(1,2,3,4)
print("Numpy Matrix:\n", np_mat)
print("\nTorch Matrix:", tc_mat)

torch.set_printoptions(precision=5) # change print options, just link numpy
print("\nTorch Matrix:", tc_mat)
torch.set_printoptions(profile='default') # back to default_pretty_printing
# print("\nTorch Matrix:", tc_mat)
Numpy Matrix:
 [[[[-1.26951553 -0.5783194  -0.2817905  -1.91081633]
   [-1.03255087  0.09356858  0.87141257  0.24527875]
   [-1.57634325 -0.95006681 -0.54418479 -0.81721992]]

  [[ 1.52620998  1.19114026  0.12362855  0.30176001]
   [-0.18951867  0.33470665 -1.13370011  1.17999206]
   [ 0.17942397 -0.15441461  0.53326092 -0.22955392]]]]

Torch Matrix: 
(0 ,0 ,.,.) = 
  1.6238  0.4029 -1.6028  0.4393
  0.0485 -0.5608 -1.3842  1.6696
  0.9837  0.7565 -0.8661 -0.6509

(0 ,1 ,.,.) = 
 -0.3699 -0.8046  0.3459  0.3806
  0.0225 -0.0216  0.5001  0.4924
  0.2519 -1.1100 -1.5480 -0.0549
[torch.FloatTensor of size 1x2x3x4]


Torch Matrix: 
(0 ,0 ,.,.) = 
 1.62381 0.40293 -1.60275 0.43934
 0.04847 -0.56078 -1.38419 1.66964
 0.98369 0.75650 -0.86610 -0.65095

(0 ,1 ,.,.) = 
 -0.36994 -0.80465 0.34593 0.38057
 0.02247 -0.02161 0.50013 0.49238
 0.25189 -1.10998 -1.54797 -0.05489
[torch.FloatTensor of size 1x2x3x4]

Easy defination and assignment with numpy

  • torch.from_numpy() creates a Tensor from a numpy.ndarray.

The returned tensor and ndarray share the same memory. Modifications to the tensor will be reflected in the ndarray and vice versa.
The returned tensor is not resizable.

1
2
3
tc_mat = torch.from_numpy(np_mat)
print("Numpy Matrix:\n", np_mat)
print("\nTorch matrix from Numpy:", tc_mat)
Numpy Matrix:
 [[[[ 0.28134406 -0.34748676  0.83155334  0.85618986]
   [-0.95961146 -0.68374939 -1.67800331  0.84947823]
   [-0.72478517  0.71535117 -2.02988345 -0.1911564 ]]

  [[ 1.04884339  0.35382358 -0.69535152  0.27244267]
   [-0.18157492 -0.02892887 -0.54348221  1.49079913]
   [ 0.6273026   0.86512992 -1.02024843 -0.58441433]]]]
Torch matrix from Numpy:
 
(0 ,0 ,.,.) = 
  0.2813 -0.3475  0.8316  0.8562
 -0.9596 -0.6837 -1.6780  0.8495
 -0.7248  0.7154 -2.0299 -0.1912

(0 ,1 ,.,.) = 
  1.0488  0.3538 -0.6954  0.2724
 -0.1816 -0.0289 -0.5435  1.4908
  0.6273  0.8651 -1.0202 -0.5844
[torch.DoubleTensor of size 1x2x3x4]

Data preprocessing

  • torch.clamp(x, min, max)
  • torch.cat((x, x, …, x), 0)
1
2
3
4
5
6
7
a = torch.randn(3,4)
print('Original Matrix a:', a)
print('\nAfter torch.clamp:', torch.clamp(a, min=-0.5, max=0.5))

cat0, cat1 = torch.cat((a, a, a), 0), torch.cat((a, a), 1)
print('\nAfter torch.cat((a,a), axis=1):', cat1)
print('\nAfter torch.cat((a,a,a), axis=0):', cat0)
Original Matrix a: 
-0.0708  0.7695  0.4194 -0.2349
-1.6515  0.4604  0.7691  0.3826
-0.7259 -0.0289 -0.7962  0.5046
[torch.FloatTensor of size 3x4]


After torch.clamp: 
-0.0708  0.5000  0.4194 -0.2349
-0.5000  0.4604  0.5000  0.3826
-0.5000 -0.0289 -0.5000  0.5000
[torch.FloatTensor of size 3x4]


After torch.cat((a,a), axis=1): 
-0.0708  0.7695  0.4194 -0.2349 -0.0708  0.7695  0.4194 -0.2349
-1.6515  0.4604  0.7691  0.3826 -1.6515  0.4604  0.7691  0.3826
-0.7259 -0.0289 -0.7962  0.5046 -0.7259 -0.0289 -0.7962  0.5046
[torch.FloatTensor of size 3x8]


After torch.cat((a,a,a), axis=0): 
-0.0708  0.7695  0.4194 -0.2349
-1.6515  0.4604  0.7691  0.3826
-0.7259 -0.0289 -0.7962  0.5046
-0.0708  0.7695  0.4194 -0.2349
-1.6515  0.4604  0.7691  0.3826
-0.7259 -0.0289 -0.7962  0.5046
-0.0708  0.7695  0.4194 -0.2349
-1.6515  0.4604  0.7691  0.3826
-0.7259 -0.0289 -0.7962  0.5046
[torch.FloatTensor of size 9x4]

Data Select

  • torch.masked_select
  • torch.index_select

    Returned Tensor does not use the same storage as the original Tensor
    and masked_select returns a 1-dim tensor

1
2
3
4
5
6
7
8
9
10
11
12
select = torch.LongTensor([0, 2])
sel0 = torch.index_select(a, 0, select)
sel1 = torch.index_select(a, 1, select)
print('Original Matrix a:', a)

mask = a.gt(0.5) # if element is greater than 0.5
print('\nMask for greater_than_0.5:', mask)
print('\nAfter torch.masked_select(a, mask):', torch.masked_select(a, mask))

print('We select index: select = torch.LongTensor([0, 2])')
print('\nAfter torch.index_select(a, axis=0, select):', sel0)
print('\nAfter torch.index_select(a, axis=1, select):', sel1)
Original Matrix a: 
-0.0708  0.7695  0.4194 -0.2349
-1.6515  0.4604  0.7691  0.3826
-0.7259 -0.0289 -0.7962  0.5046
[torch.FloatTensor of size 3x4]


Mask for greater_than_0.5: 
 0  1  0  0
 0  0  1  0
 0  0  0  1
[torch.ByteTensor of size 3x4]


After torch.masked_select(a, mask): 
 0.7695
 0.7691
 0.5046
[torch.FloatTensor of size 3]

We select index: select = torch.LongTensor([0, 2])

After torch.index_select(a, axis=0, select): 
-0.0708  0.7695  0.4194 -0.2349
-0.7259 -0.0289 -0.7962  0.5046
[torch.FloatTensor of size 2x4]


After torch.index_select(a, axis=1, select): 
-0.0708  0.4194
-1.6515  0.7691
-0.7259 -0.7962
[torch.FloatTensor of size 3x2]

Data Squeeze

  • torch.squeeze

You can select which axis to be squeezed

1
2
3
4
5
z = torch.zeros(4,5,1,2)
print("Size of Z: \t ", z.size())
print("Size of Z after squeeze(z):\t ", torch.squeeze(z).size())
print("Size of Z after squeeze(z, 1):\t ", torch.squeeze(z, 1).size())
print("Size of Z after squeeze(z, 2):\t ", torch.squeeze(z, 2).size())
Size of Z:                       torch.Size([4, 5, 1, 2])
Size of Z after squeeze(z):      torch.Size([4, 5, 2])
Size of Z after squeeze(z, 1):      torch.Size([4, 5, 1, 2])
Size of Z after squeeze(z, 2):      torch.Size([4, 5, 2])
  • Serialization
  • Parallelism
    • torch.set_num_threads(int)
    • torch.get_num_threads() → int
  • Check Matrix Comparison
    • torch.eq()/ge()/gt() -> Tensor
    • torch.equal() -> Bool

In-place random sampling

There are a few more in-place random sampling functions defined on Tensors as well. Click through to refer to their documentation:

Distribution description
torch.Tensor.bernoulli_() in-place version of torch.bernoulli()
torch.Tensor.cauchy_() numbers drawn from the Cauchy distribution
torch.Tensor.exponential_() numbers drawn from the exponential distribution
torch.Tensor.geometric_() elements drawn from the geometric distribution
torch.Tensor.log_normal_() samples from the log-normal distribution
torch.Tensor.normal_() in-place version of torch.normal()
torch.Tensor.random_() numbers sampled from the discrete uniform distribution
torch.Tensor.uniform_() numbers sampled from the uniform distribution

Variable for requires_grad or not

1
2
3
4
5
6
7
8
9
from torch.autograd import Variable
# http://pytorch.org/docs/master/autograd.html#torch.autograd.Function
x = Variable(torch.randn(5, 5))
y = Variable(torch.randn(5, 5))
z = Variable(torch.randn(5, 5), requires_grad=True)
a = x + y
b = a + z
print("Variable a=x+y requires grad: ", a.requires_grad)
print("Variable b=a+z requires grad: ", b.requires_grad)
Variable a=x+y requires grad:  False
Variable b=a+z requires grad:  True

Finished Basic

I can’t wait for Machine learning!
Let’s start new learning into torch.nn~

dynamic_graph