Bootstrap the library

[cwognum]

As @zhu0619 and me were working through polaris-hub/polaris#98, we realized that the curation module within the Polaris library really should be a stand-alone package. It is independent from the broader Polaris codebase. Up until now this was fine, but as we go public with Polaris, we will need to be more careful about our release strategy, which would unnecessarily slow down the development of the curation module.

This PR moves the curation module into its own package, temporarily named alchemy.

Refactoring

As part of this PR, we decided to refactor the code base to be better setup for future maintenance.

In summary:

• We added a Curator class. This class specifies the curation process as a number of steps. It can be serialized to be easily saved to and loaded from JSON. The goal with this is to make the process more easily reproducible.
• We reimplemented all the curation steps as individual, serializable BaseAction objects.
• We then added the curation.functional module, inspired by torch.nn.functional, to easily use any of the curation steps in isolation, outside of the object-based approach.
• We added a CurationReport, which is produced by the Curator and holds relevant information about the curation process (for now just logs and images) and can be exported to different formats through Broadcasters, such as the LoggerBroadcaster and HTMLBroadcaster.
• We added the visualization module and refactored the visualizations to be more general.

The object-based API now looks like:

# Define the curation workflow
curator = Curator(
    steps=[
        MoleculeCuration(input_column="smiles"),
        OutlierDetection(method="zscore", columns=["SOL"]),
        Discretization(input_column="SOL", thresholds=[-3]),
    ],
    parallelized_kwargs = { "n_jobs": -1 }
)

# Run the curation
dataset, report = curator(dataset)

You could do something similar with the functional API, e.g.:

y = np.random.normal(0, 1, 1000)
is_outlier = detect_outliers(y)
visualize_distribution_with_outliers(y, is_outlier)

Relation to Polaris Recipes

Now that we have this system, I'm thinking we should require every polaris-recipe to provide the serialized Curator object as JSON. That way, everyone that has access to the base dataset could reproduce the process with the CLI we added.

TODO

(Note: We don't need to do all these things in this PR, but just for future reference)

• Decide on a name
• Extensively test by porting the code in the Polaris Recipes repo
• Add more documentation (bit of an empty shell right now)
• Make public and release officially!