Astrolabe

Astrolabe is a visual analytics system that reframes tabular model interpretation as an interactive feature relationship network. It bridges the gap between granular SHAP-based attributions and complex structural dependencies to uncover insights often missed by individual feature rankings.

This work is currently under review (VIS 2026 Short Paper).

Installation

git clone https://github.com/kyungjini/astrolabe
cd astrolabe
pip install .

Python: 3.10+

Quick start

import astrolabe

astro = astrolabe.model(df_X, df_y, task="regression")
astro.explore()

Save/load analysis objects:

astro.save("run.astrolabe")
reloaded = astrolabe.load("run.astrolabe")

API

• astrolabe.model(df_X, df_y, task="regression", **kwargs)
• astrolabe.load_model(df_X, df_y, task="regression", model=..., **kwargs)
• astrolabe.load(path)
• astrolabe.config(...)

Main object methods:

• Astrolabe.explore()
• Astrolabe.save(path)
• Astrolabe.plot(...) (optional plot-only helper)
• Astrolabe.status()
• Astrolabe.show_multicollinearity(threshold=0.7)

Citation

N/A

License

Apache License 2.0. See LICENSE.