pycm

Why we included this project

Accuracy alone rarely tells you where a classifier goes wrong, and PyCM targets that gap directly. It is a multi-class confusion matrix library that accepts either predicted-versus-actual vectors or a pre-built confusion table, then returns per-class and overall metrics such as precision, recall, F1, macro-aggregated scores, kappa, and AUC-based interpretation, plus normalized tables and plots when Matplotlib or Seaborn are available. That makes it a solid choice for research pipelines, Kaggle-style experiments, and teams that want reproducible model evaluation without hand-writing their own metric code. Because it also takes a directly supplied matrix, it works when you only have an existing table from another system, so it stays useful as a general utility rather than something tied to one framework.