More Efficient Backend

• Major feature: Refactor to no longer copy arrays
• Major feature: Rust calls to keep GIL (side-effect of not copying data)
• Efficiency: Single-threaded performance has improved in majority of cases (no copying)
• Minor: Add bump2version usage

Adding `mypy` support and `bandit` security scans

Adding mypy support and bandit security scans

Adding IoU

Adds iou function.

Additional Helper Functions

Adding high-level API including binary_tp_fp_fn, binary_stats and stats functions to increase functionality.

Main Release

Supports:

• binary precision, recall and f1-score
• multi-class precision, recall and f1-score
• calculation of confusion matrices
• benchmarks folder shows both speedup over scikit-learn, numpy (in binary case), and a multithreading example

Adding Confusion Matrix

Adds calculation of confusion matrix to fast-stats. See benchmarks/timeit.ipynb for more details.

Updated CICD

Enhanced README, updated CICD, better PyPi metadata, and badges!

Faster binary implementations

Still only supporting binary calculations for precision, recall, and f1. For smaller data types like: bool, uint8, etc. the methods are faster than the pure numpy equivalents. However for large data types, the methods can be slower.

Next:

• Fast confusion matrix generation.

Initial Release w/ support for binary arrays

Limited project setup and experimenting w/ GitHub actions still.