|
|
--- |
|
|
license: mit |
|
|
library_name: generic |
|
|
pipeline_tag: tabular-classification |
|
|
tags: |
|
|
- fairness |
|
|
- calibration |
|
|
- multicalibration |
|
|
- gradient-boosting |
|
|
- gbdt |
|
|
- decision-trees |
|
|
- trustworthy-ai |
|
|
- tabular |
|
|
- risk-assessment |
|
|
- risk |
|
|
arxiv: "2509.19884" |
|
|
model-index: |
|
|
- name: MCGrad |
|
|
results: [] |
|
|
--- |
|
|
|
|
|
# MCGrad: Multicalibration at Web Scale |
|
|
**Production-ready multicalibration for machine learning.** *Developed by Meta. Accepted at KDD 2026.* |
|
|
|
|
|
**Paper:** [arXiv:2509.19884](https://arxiv.org/abs/2509.19884) |
|
|
**Official Code:** [github.com/facebookincubator/MCGrad](https://github.com/facebookincubator/MCGrad) |
|
|
**Documentation:** [mcgrad.dev](https://mcgrad.dev) |
|
|
|
|
|
## Overview |
|
|
MCGrad is a library for production-ready multicalibration. It ensures your ML model predictions are well-calibrated not just globally, but across virtually any segment defined by your features. |
|
|
|
|
|
## Installation |
|
|
```bash |
|
|
pip install mcgrad |
|
|
``` |
|
|
## Getting started |
|
|
See [mcgrad.dev](https://mcgrad.dev) |
|
|
|
|
|
|
