v0.3.0a3

Added

• explainability_tokenizer as an optional argument to the add_model method, to be used by the explainability techniques.

Changed

• requirements_txt_file no longer optional for model uploads.
• Remove id from POST params to API server.

Fixed

• Added protobuf==3.2.0 to requirements to fix bug with model deployment.

v0.3.0a1

Changed

• Default Unbox server URL (https://api-staging.unbox.ai/).

v0.3.0a2

Fixed

• Fixed link to project page when loading / creating a project.
• Presigned url endpoint when using AWS / GCP / Azure.

Changed

• Removed links when uploading dataset and models. Just the project link is appropriate.

v0.3.0a0

Added

• A Project helper class.
• A convenience method create_or_load_project which loads in a project in if it is already created.

Changed

• Models and datasets must be added to projects.
• Deprecates categorical_features_map in favor of categorical_feature_names for model and dataset uploads.
• Moved TaskType attribute from the Model level to the Project level. Creating a Project now requires specifying the TaskType.
• Removed name from add_dataset.
• Changed description to commit_message from add_dataset, add_dataframe and add_model.

v0.2.0a1

Fixed

• Fail early if custom_model_code, dependent_dir or requirements_txt_file are None when model type is ModelType.custom.
• Fail early if model is not None when model type is ModelType.custom.

v0.2.0a0

Added

• Accepts AZURE as a DeploymentType.

Changed

• Compatibility with Unbox backend storage and data refactor.

v0.1.2

Fixed

• Default Unbox server URL (https://api.unbox.ai/).

v0.1.1

Added

• Can specify ModelType.keras.
• API documentation.

Changed

• Parameter ordering for add_model, add_dataset, add_dataframe for clarity's sake.

Fixed

• Fix bug when predict function is a list with numpy objects.
• Better error message if return type isn't a list or numpy array.
• Prevents dataset upload when label_column_name is also in feature_names.