pytorch_learning

learning about how to use pytorch

PyTorch is a deep learning framework for fast, flexible experimentation

Documents

English ver: https://pytorch.org/docs/stable/index.html

chinese ver: https://pytorch-cn.readthedocs.io/zh/latest/
      https://ptorch.com/docs/1/
      index

Installtion

Base on python3

• Windows

  • Step1: Install ANACONDA: https://www.anaconda.com/download/

    Add anaconda to environment path:
    installpath\Anaconda3
    installpath\Anaconda3\Scripts
    installpath\Anaconda3\Library\bin

  • Step2: Install CUDA: https://developer.nvidia.com/cuda-downloads

  • Step3: Create a virtural environment for pytorch

    conda create -n pytorch python=3.6

  • Step4: Activate your virtural environment

    activate pytorch

    PS: if you finish your work in the virtural environment and want to quit the ve, use the flowing command:

    deactivate pytorch
    

  • Step5: In the virtural environment:

    # for cuda8.0
    conda install pytorch cuda80 -c pytorch 
    # for cuda9.0
    conda install pytorch cuda90 -c pytorch 

  • Step6: use the flowing python code to test:

    import torch
    print(torch.__version__)

  • Step7: install torchvision

    pip install torchvision

• Linux

  • Step1: install Anaconda

  • Step2: Add Tsinghua Open Source Mirror

    conda config

    Use the above command to generate a configuration file .condarc. Then

    conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
    conda config --set show_channel_urls yes

    delete '-defaults', the content of .condarc just like:

    channels:
      - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
      - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/
      - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/msys2/
    show_channel_urls: true

  • Step3: install pytorch and torchvision:

    # for cuda8.0
    conda install pytorch torchvision -c pytorch
    # for cuda9.0
    conda install pytorch torchvision cuda90 -c pytorch

Models

The directory models/ has common deep learnning network models that were implemented by pytorch:

alexnet
VGG
ResNet
DenseNet
BNinception
caffe_resnet
fb_resnet
inception_resnetv2
inceptionv4
nasnet
nasnet_mobile
pnasnet
polynet
vggm
wide_resnet
xception.

ps：由于模型最后没有经过softmax，所以多分类下loss function请使用CrossEntropyLoss。假如修改了代码让模型最后的输出经过了softmax，那么loss function使用NLLLoss(CrossEntropyLoss 等价于 softmax + NLLLoss)

How to load dataset

关于读入的数据集可以有三种方法制作，详情点这里

How to train and test

关于如何训练，如何测试，如何上GPU，点这里

How to store and load model

关于如何保存训练好的模型、加载模型 以及 fine-tune 点这里

pytorch版本升级带来的变化

0.3.1 -> 0.4

Example

The direcoty examples/ has simple examples for using pytorch

mnist.py: A simple cnn to train and test mnist datasets

FashionAI: 阿里天池全球挑战赛-服饰属性标签识别

To be add