Doing exercises from the book (v2) & playing around
Official reference notebook: https://github.com/ageron/handson-ml2
If using a virtualenv, need to explicitly add it to the Jupyter kernels: python -m ipykernel install --user --name=XXX where XXX is the name of the virtualenv
This ensures it uses the packages in the virtualenv, and not the system python version (check with plt.__file__)
You can list the jupyter kernels: jupyter kernelspec list
Use nbdev package to do conversion: https://www.fast.ai/2020/01/20/nb2md/
Specifically: nbdev_nb2md xxxx.ipynb to create: xxxx.md and xxxx_files. These can be copied to the website repository.
The settings.ini file is required for nbdev_nb2md.
On my Mac, pip needed to compile XGBoost, rather than using the prebuilt wheels. This will error with:
Could NOT find OpenMP_C (missing: OpenMP_C_FLAGS OpenMP_C_LIB_NAMES)
clang on a Mac is missing the OpenMP libraries, so let's install it first: brew install libomp (you will also need cmake)
pip install xgboost should now work fine.
(Note to self: to avoid updating all other homebrew packages can do brew install --ignore-dependencies libomp or HOMEBREW_NO_AUTO_UPDATE=1 brew install libomp)
Maybe also see http://stechschulte.net/2016/03/20/openmp-osx-cmake.html, although I didn't follow those steps.