Dataset api

[chrisiacovella]

Pull Request Summary

This adds in a new module called "curate" that provides an API for dataset curation.

The general hierarchy is that each dataset contains records, and each record contains properties. In this, properties are defined using pydantic models (e.g., AtomicNumbers, Energy, Forces, PartialCharge, etc.) . The pydantic property models ensure that for each value we ensure: units, property type (e.g., length, energy, force), and classification (per_atom, per_system), and also validates that the shape of the inputted array matches the expectation related to the classification. Records collect the properties and validate that consistent number of configurations and atoms exist across properties. The dataset class provides functions to write to an hdf5 file, converting units to the specified unit system.

The dataset also can be initialize in an "append" mode whereby properties associated with individual configurations can be added separately, and the code will automatically append to the internal numpy array (performing any unit conversion necessary).

This PR is still a WIP, as it will require also converting old curation scripts to use the new api, and to revise the dataset class in model forget to accept the new format (minor changes to the terminology we use for classification).

Key changes

Notable points that this PR has either accomplished or will accomplish.

• Change 1

Questions

• Question 1

Associated Issue(s)

• Dataset API #321

Pull Request Checklist

• Issue(s) raised/addressed and linked
• Includes appropriate unit test(s)
• Appropriate docstring(s) added/updated
• Appropriate .rst doc file(s) added/updated
• PR is ready for review

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 79.07%. Comparing base (f013190) to head (a581f5c).

Additional details and impacted files

[chrisiacovella]

The NPZ file that is generated from the dataset is somewhat dynamic (and we can't just look at the md5 checksum). To see if the code needs to regenerate that npz file or just use the one already that exists, we look at a metadata file that is written out (as a json file). This include the md5checksum of the hdf5 file and also the properties of interest that were used to generate the npz file. We need to also put the element filter in there and look at it. If the element filter has changed, we need to regenerate that npz file.