Metadata-Version: 2.2
Name: dataore
Version: 0.11.1
Summary: Lightweight data mining toolkit: profiling, sketches, diffing, feature engineering, ML models (trees, boosting, SVM, clustering, feature selection), dimensionality reduction (PCA/LDA/MDS/Isomap/t-SNE), and ASCII visualization (scatter/histogram/heatmap/boxplot/line)
Author: Equinox
License: MIT
Requires-Python: >=3.10
Description-Content-Type: text/markdown
Requires-Dist: pandas>=1.5
Requires-Dist: numpy>=1.23
Requires-Dist: scipy>=1.9

# DataOre

Lightweight data mining and machine-learning tools built on NumPy, pandas and SciPy.

## 0.11.1 correctness fixes

- ROC-AUC rejects non-finite prediction scores instead of stalling on NaN.
- AutoML fits scaling separately within each training fold and on the final
  training split; validation and holdout data no longer determine scaling.
- Failed or non-finite model evaluations cannot win model selection. Their
  errors remain available in the results and leaderboard.
- Regression leaderboards rank lower MSE first; prediction reuses the selected
  model's training statistics.
