(documentation) pip full install, numpy, and scikit-learn

[draeath]

In the README, the pip installation method is show as such with the assumption that this is a one-stop-shop command that will pull everything necessary in.

pip install pandashells[full]

However, in doing so, scikit-learn will fail to install if numpy is not already installed. This has explicitly been left unaddressed by scikit-learn as a wontfix. This is manifest as an error message while building the wheels, with the text ModuleNotFoundError: No module named 'numpy'

Numpy must be installed (either by a package manager (apparently preferred), or pip) prior to running the command to install pandashells[full].

If you do run as written currently, the install completes, leaving scikit-learn broken but seemingly installed. This can be repaired as such, after installing numpy:

pip install --upgrade --force-reinstall --no-deps scikit-learn

The above will have pip re-perform the installation without also chaining up and redoing dependencies.

... you may also need to address scipy as well, though in my case while I had errors the first-round for both, installing just numpy seems to have resolved it.

[robdmc]

Sorry for the delayed response. I've been pretty busy of late. Thanks for bringing this up. I'll spend some time over the next few weeks coming up with a fix for this. I would love it if there could still be a "one stop shop" install command. If you have any suggestions on how to pull this off, I'd be appreciative.