First Public Release

Since 0.1.0, the package has seen a lot of additions, including:

• A new docs site.
• bavard_ml_utils.ml.conv_graph.ConvGraph, which allows a graph to be built out of a chatbot conversations dataset. The graph can be used for analysis, dataset augmentation, and computing "soft" metrics. See "Conversation Graph: Data Augmentation, Training, and Evaluation for Non-Deterministic Dialogue Management" by Gritta et al. 2021 for more details.
• The bavard_ml_utils.persistence sub-package, which includes features for easily persisting Pydantic objects, as well as versioned artifacts produced by machine learning models. Supports Pydantic objects with numpy.ndarray fields out of the box.
• And more!

New Data Model Interfaces & LabeledDataset Class

• The Bavard in-memory data model interfaces in bavard_ml_common.types have been refactored to be in harmony with the latest.
• A new bavard_ml_common.ml.dataset.LabeledDataset abstract class has been added which can allow an arbitrarily-typed labeled list acting as a dataset to inherit nice behavior for things like cross validation, train/test splitting, and upsampling to balance by label.

Patch Release

• Updated dependencies
• New deep learning framework agnostic (pure numpy) one-hot utility. Can be used for arrays of arbitrary dimensionality.