flow-forecast

This release adds support for the following:

• Focal Loss function for imbalanced classification and anomaly detection.
• Allows training using multiple criterion (experimental support)
• Fixes bug related to training the Informer model

We are aiming for NYE release of Flow Forecast 1.0.

This (pre)-release accomplishes the following items

• Fixes bug in the Informer decoding process
• Adds support for time series classification
• Creates docstrings and additional ReadTheDocs documentation

We are very close to release 1.0 which should be out before October.

Change Log

• Support for using SHAP with the Informer in CPU mode
• Fixes several bugs related to handling of Informer info
• Increased GPU utilization for specific models
• Fix for a PyPi related versioning error.
• Fixes a bug related to Wandb and SHAP.

Coming in version 0.99

• GPU support for Informer with SHAP
• Gaussian Loss bug fixes
• Addition of series_id for core data loaders.

Fix an issue related to dependencies and allow for the using of a single config file. Also gets rid of requiring batch_size in the config. No long config dataset_params.

• Informer support for SHAP will be in next release

This release adds support for Informer models.

In this release we add several important bug fixes and new features.

This release includes general updates to dependencies and some inference mode support.