Adding OneHotEncoder HypergridAdapter

[edcthayer]

OneHotEncoder Hypergrid Adapter allows categorical dimensions to be transformed to dummy variables using the sklearn OneHotEncoder. By default, each categorical dimension will produce its own collection of OneHotEncoded dummy variables, but one may also generate a OneHotEncoding of the cross product of all categorical dimension categories using the merge_all_categorical_dimensions argument (as used in the RERF surrogate model).

Once this adapter is approved, it will be integrated into the regression-enhanced random forest regression (RERF) model.

[byte-sculptor]

Would it be good to check if projected_df[column].isin([0,1]).all() instead of between(0, 1)?

[bpkroth]

@amueller to comment more, but I think these are supposed to not have the space separation in order to be processed by sphinx for API publication.

[bpkroth]

Could see also providing access to the the map for lookup instead of just the array of names, but that could always be added later when the need arises.

[bpkroth]

Suggested change

	hence that list is augmented to include the string 'nan' which pandas.DataFrame.apply(map(str)) will produce from a np.NaN value.
	hence that list is augmented to include the string 'nan' which pandas.DataFrame.apply(map(str)) will produce from a np.NaN value.

?

[bpkroth]

Similar below.

[bpkroth]

I don't have any substantial feedback.
I think there are some comment styling things that will show up when we try to publish the API docs here: https://microsoft.github.io/MLOS/python_api/
I don't think we have instructions written up on testing that yourself yet.
The rough outline goes something like this:

cd website
# generate the Python APIs to be served at http://localhost:8080/MLOS/python_api/
make sphinx-site # repeat this one as necessary while the docker container is running
# generate the rest of the http://localhost:8080/MLOS site content
make hugo-site # optional
# start a local webserver
# borrowed from https://github.com/microsoft/MLOS/blob/main/website/test_site_links.sh#L48
docker run -d --name mlos-website-link-checker -v $PWD:/src/MLOS/website -v $PWD/nginx.conf:/etc/nginx/conf.d/mlos.conf -p 8080:8080 nginx:latest
# now point your browser at http://localhost:8080/MLOS/python_api/ to double check things

Note: the instructions assume a Linux environment atm (e.g. via WSL2)

[bpkroth]

I don't have any substantial feedback.
I think there are some comment styling things that will show up when we try to publish the API docs here: https://microsoft.github.io/MLOS/python_api/
I don't think we have instructions written up on testing that yourself yet.
The rough outline goes something like this:

cd website
# generate the Python APIs to be served at http://localhost:8080/MLOS/python_api/
make sphinx-site # repeat this one as necessary while the docker container is running
# generate the rest of the http://localhost:8080/MLOS site content
make hugo-site # optional
# start a local webserver
# borrowed from https://github.com/microsoft/MLOS/blob/main/website/test_site_links.sh#L48
docker run -d --name mlos-website-link-checker -v $PWD:/src/MLOS/website -v $PWD/nginx.conf:/etc/nginx/conf.d/mlos.conf -p 8080:8080 nginx:latest
# now point your browser at http://localhost:8080/MLOS/python_api/ to double check things

Note: the instructions assume a Linux environment atm (e.g. via WSL2)

Actually, now that I look at it, we seem to have neglected to transfer some API doc comments to some of the public calls after we added the _ underscore prefix.

For example:
https://microsoft.github.io/MLOS/python_api/api/mlos.Spaces.HypergridAdapters.HypergridAdapter.html
https://github.com/Microsoft/MLOS/blob/732b9d7/source/Mlos.Python/mlos/Spaces/HypergridAdapters/HypergridAdapter.py#L98

We should fix that, though I don't think it needs to be done in this PR.