python环境配置

python相关环境配置：Miniconda，PyTroch，Jupyter， venv。

Miniconda

Miniconda是一个轻量级的Conda包管理器

安装包

所有安装包：repo.anaconda.com/miniconda

Linux: python3.8 Miniconda3 Linux 64-bit

​ python3.7 Miniconda3 Linux 32-bit

# 以前的版本没有ssh问题
wget https://repo.anaconda.com/miniconda/Miniconda3-4.7.12.1-Linux-x86_64.sh

配置

初始化终端：

~/miniconda3/bin/conda init

如果失败可以手动添加bin目录。

echo "export PATH=\$PATH:/home/pi/miniconda3/bin" >> .bashrc

使用

新建环境

# 新建名称为test的环境
conda create --name test python=3.8 -y

# arm只能创建python=3.4的环境
conda create -n test python=3.4 -y

查看所有的环境

conda env list

使用环境

# 激活环境
conda activate test
source activate test # arm

# 退出当前环境
conda deactivate
source deactivate # arm

# 删除环境
conda remove -n test1 --all

rpi

armv7l: miniconda3-latest-linux-armv7l.sh

Berryconda3-2.0.0-Linux-armv7l.sh

arm版本的miniconda最高只能安装python3.4，如果需要安装更高版本的python，需要第三方conda，这里使用的是 berryconda，目前最高支持到python3.6。

conda config --add channels rpi

conda create --name test python=3.6 -y

venv

python自带的虚拟环境模块

# 新建
python -m venv /path/venv_name

# 激活
source /path/venv_name/bin/activate

# 退出
deactivate

https://www.python.org/ftp/python/3.8.5/Python-3.8.5.tgz

Pytorch

old version lib torch: https://github.com/pytorch/pytorch/issues/40961

Pytorch，选择对应cuda版本。

torch.version # PyTorch version
torch.cuda.is_available()
torch.version.cuda # Corresponding CUDA version
torch.backends.cudnn.version() # Corresponding cuDNN version
torch.cuda.get_device_name(0) # GPU type

build

# install pytorch and dependences
git clone --depth 1 --recurse-submodule https://github.com/pytorch/pytorch.git
conda create -y --name pytorch-build python=3.8
conda activate pytorch-build
conda install -y astunparse numpy ninja pyyaml mkl mkl-include setuptools cmake cffi typing_extensions future six requests dataclasses pkg-config libuv

# arm64
mkdir pytorch-build-arm64
cd pytorch-build-arm64
cmake -DBUILD_SHARED_LIBS:BOOL=ON -DCMAKE_OSX_ARCHITECTURES=arm64 DCMAKE_OSX_DEPLOYMENT_TARGET=12.10 -DUSE_MKLDNN=OFF -DUSE_QNNPACK=OFF -DUSE_PYTORCH_QNNPACK=OFF -DBUILD_TEST=OFF -DUSE_NNPACK=OFF -DCMAKE_BUILD_TYPE:STRING=Release -DPYTHON_EXECUTABLE:PATH=`which python3` -DCMAKE_INSTALL_PREFIX:PATH=../pytorch-install-arm64 ../pytorch
cmake --build . --target install
cd ..

# x86_64
mkdir pytorch-build-x86_64
cd pytorch-build-x86_64
cmake -DBUILD_SHARED_LIBS:BOOL=ON -DCMAKE_OSX_ARCHITECTURES=x86_64 -DCMAKE_OSX_DEPLOYMENT_TARGET=12.10 -DUSE_MKLDNN=OFF -DUSE_QNNPACK=OFF -DUSE_PYTORCH_QNNPACK=OFF -DBUILD_TEST=OFF -DUSE_NNPACK=OFF -DCMAKE_BUILD_TYPE:STRING=Release -DPYTHON_EXECUTABLE:PATH=`which python3` -DCMAKE_INSTALL_PREFIX:PATH=../pytorch-install-x86_64 ../pytorch
cmake --build . --target install
cd ..

test

# CMakeLists.txt
cmake_minimum_required(VERSION 3.0 FATAL_ERROR)
project(example-app)

set(Torch_DIR /Users/sanzo/Software/pytorch-install-arm64)
set(CMAKE_PREFIX_PATH "/Users/sanzo/Software/pytorch-install-arm64/share/cmake/Torch")
find_package(Torch REQUIRED)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${TORCH_CXX_FLAGS}")

add_executable(example-app example-app.cpp)
target_link_libraries(example-app "${TORCH_LIBRARIES}")
set_property(TARGET example-app PROPERTY CXX_STANDARD 14)

#include <torch/torch.h>
#include <iostream>

int main() {
  std::cout << "PyTorch version: "
    << TORCH_VERSION_MAJOR << "."
    << TORCH_VERSION_MINOR << "."
    << TORCH_VERSION_PATCH << std::endl;
  torch::Tensor tensor = torch::rand({2, 3});
  std::cout << tensor << std::endl;
}

/*
 0.0598  0.7058  0.0322
 0.2230  0.4112  0.9342
[ CPUFloatType{2,3} ]
*/

references

old version lib torch: https://github.com/pytorch/pytorch/issues/40961

https://dev-discuss.pytorch.org/t/universal-binaries-for-libtorch-mac/229

https://github.com/pytorch/pytorch/issues/63558

jupyter notebook

ipykernel

通过ipykernel管理jupyter notebook的内核。

# 激活环境
activate env_name

# 安装ipykernel
pip install ipykernel

# 添加kernel
python -m ipykernel install --name env_name

# 删除内核
jupyter kernelspec remove kernelname

# 查看所有内核
jupyter kernelspec list

远程访问

生成密钥

# 通过ipython生成密码
ipython

In [1]: from notebook.auth import passwd
In [2]: passwd()
Enter password: 
Verify password: 
Out[3]: 'xxxxxxxxxxxxxxxxxxxxxx'

# 生成配置文件，并添加如下配置
jupyter notebook --generate-config

c.NotebookApp.ip='0.0.0.0'
c.NotebookApp.password = u'xxxxxxxxxxxxxxxxxxxxxx'
c.NotebookApp.open_browser = False
c.NotebookApp.port =8888

然后就可以通过https://ip:8888远程访问jupyter notebook