# uber/manifold

A model-agnostic visual debugging tool for machine learning

Repository: https://github.com/uber/manifold
Canonical: https://ross.abutalabs.com/products/uber-manifold
Language: JavaScript
License: Apache-2.0
License Family: permissive
Topics: incubation, machine-learning, visualization
Last push: 2025-02-05T23:02:02+00:00

## Health v2 (maintenance only)
Score: 25/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 5, release rhythm 8, longevity 100
- inputs: {"age_days": 2763, "days_push": 574, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1672, forks 116 (observed 2026-08-28T04:05:20.159184+00:00)

## What it is
Manifold is a model-agnostic visual debugging tool for machine learning from Uber. It helps ML practitioners identify which subsets of data a model predicts poorly and explains potential causes by visualizing feature distribution differences between better- and worse-performing segments.

## Use cases
- debug why my ML model performs poorly on certain data subsets
- visualize feature distributions across model performance segments
- compare predictions from multiple models on an evaluation dataset
- find which data slices a classifier gets wrong
- explain model errors beyond summary metrics like AUC or RMSE
- embed an ML performance visualization component in my own web app
- upload a CSV of predictions and ground truth to analyze model behavior

## When to choose
- you need to visually diagnose which data subsets a model mispredicts
- you want a model-agnostic tool that works with any classifier's predictions
- you want to embed an interactive ML debugging visualization in a React web app
- you have evaluation data with features, predictions, and ground truth in CSV or JS format

## When to avoid
- you need automated model training or hyperparameter tuning rather than debugging
- you need real-time monitoring of models in production
- your datasets are far larger than the recommended 10k-15k instances without sampling
- you need a non-visual, programmatic error analysis library

## Facets
- artifact type: application
- maturity: maintenance
- function: data-visualization, machine-learning, analytics
- domain: machine-learning, data-visualization, data-science
- platform: cross-platform
- tags: visual-debugging, model-agnostic, ml-debugging, uber, react-component, incubation, web, nodejs

## Member repositories
- uber/manifold (main) score 25

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:20.159184+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:42:39.345779+00:00, confidence not recorded.
  - readme: https://github.com/uber/manifold (fetched 2026-08-28T04:05:20.159184+00:00, sha b5367a03b178)
- Data as of 2026-08-30T08:39:29.467469+00:00.