pytalite 1.0.0

Python/Pyspark Package for Model-agnostic Evaluation and Diagnosis.

Pytalite: Model-agnostic diagnostics and evaluation tool for Python and Pyspark

Overview

Evaluating model performance and diagnosing models are important steps in machine learning model development. Pytalite provides a straightforward wrapper for visualizing model performance and diagnosing machine learning models. It enables users to visualize key snapshots of model performance, e.g. discrete precision-recall curves, probability density plots, model-agnostic feature importance, partial dependence plots, accumulative local effect plots, and feature correlation plots.

Requirements

Pytalite for python is developed under python 3.7, but is compatible with python 2.7. Pytalite for pyspark is developed to support spark 2.0 and above.

matplotlib ≥ 2.2.x (although 1.4.3 also works, higher version is recommended)

numpy ≥ 1.9.x

scipy ≥ 0.15.x

multiprocess ≥ 0.70.4

Package Structure

Highlights

Pytalite provides the following model evaluation and diagnostic algorithms:

• Discrete Precision/Recall Plot (Binary-Classification only)
• Feature Correlation Plot (Binary-Classification only)
• Probability Density Plot (Binary-Classification only)
• Feature Importance Plot (Binary-Classification only)
• Accumulated Local Effect Plot (Binary-Classification / Regression, numerical feature)
• Partial Dependence Plot (Binary-Classification / Regression)

Examples

See examples folder for usage examples.

Gallery

Discrete Precision/Recall Plot

---

Feature Importance Plot

---

Feature ALE Plot for numerical features