Enable Threading on Windows for PyStan 2.18+

Simple instructions to enable multithreading on Windows with clang-cl and Microsoft Visual Studio C++ build tools

These instructions need proper user privileges (administrative privileges) for installing Build Tools for Visual Studio 2017.

1. Get Build Tools for Visual Studio 2017.

Follow instructions found in https://wiki.python.org/moin/WindowsCompilers

Build Tools for Visual Studio 2017 https://visualstudio.microsoft.com/downloads/#build-tools-for-visual-studio-2017

Install the defaults and restart computer

2. Get Anaconda or Miniconda distribution

Anaconda https://www.anaconda.com/download/#windows Miniconda https://conda.io/miniconda.html

3. Create Conda virtual environment

Example uses Python 3.6, but feel free to use Python 3.7 instead.

Minimum

conda create -n stan_clang python=3.6 numpy cython clangdev -c conda-forge

Toolset I frequently use (JupyterLab + Excel tools + tools for ArviZ)

conda create -n stan_clang python=3.6 numpy cython matplotlib scipy jupyter jupyterlab notebook ipython xarray netcdf4 openpyxl xlrd pandas clangdev -c conda-forge

To list all available environments

conda env list

4. Activate environment `stan_clang`

conda activate stan_clang

For older conda distributions use

activate stan_clang

5. Get files to enable clang

Clone pystan_windows_threading

git clone https://github.com/ahartikainen/pystan_windows_threading
cd pystan_windows_threading

Add clang-cl compiler option to distutils.cfg

step_1_add_distutils_cfg.bat

Replace _msvccompiler.py & ccompiler.py with clang-cl enabled files

step_2_enable_clangcl.bat

6. Install PyStan

Minimum

pip install pystan

Install arviz for visualization

pip install pystan arviz

Done. Follow official instructions to enable multithreading on PyStan.

https://pystan.readthedocs.io/en/latest/threading_support.html

Commands

Minimum workflow

conda create -n stan_clang python=3.6 numpy cython clangdev git -c conda-forge -y
conda activate stan_clang
git clone https://github.com/ahartikainen/pystan_windows_threading
cd pystan_windows_threading
step_1_add_distutils_cfg.bat
step_2_enable_clangcl.bat
pip install pystan

Suggested workflow

conda create -n stan_clang python=3.6 numpy cython matplotlib scipy jupyter jupyterlab notebook ipython xarray netcdf4 openpyxl xlrd pandas clangdev git -c conda-forge -y
conda activate stan_clang
git clone https://github.com/ahartikainen/pystan_windows_threading
cd pystan_windows_threading
step_1_add_distutils_cfg.bat
step_2_enable_clangcl.bat
pip install pystan arviz

Create `run_all.bat` file and call it from cmd