pycaret /pycaret
Open-source, low-code AutoML platform for Python. PyCaret 4.0: sklearn-native engine + React control plane.
★ 9,863⑂ 1,837👁 138⑃ 1,113● 2,353▤ 31
+0
日增长
9,863
stars
1,837
forks
138
watchers
1,113
prs
2,353
issues
31
releases
README
PyCaret — open-source self-hosted ML platform The engine, the control plane, and the UI — all in one box. Vision · Architecture · Roadmap · Spec · Agent guide --- ⚠ 4.0 is work in progress — you're on main (the 4.0 line) PyCaret 4.0 is a ground-up architectural revamp. The 3.x line is frozen on PyPI as pycaret 3.4.0 — no further commits. Track progress in docs/revamp/STATUS.md and docs/revamp/ROADMAP.md. --- What you get A self-hosted ML platform you can run on your laptop in 5 minutes: - Engine — sklearn-1.7-based AutoML library (pycaret on PyPI). - Control plane — FastAPI backend with workspaces, projects, experiments, runs, model registry, deployments, approvals, monitoring, drift, lineage, webhooks, schedules, LLM-advisory copilots. - Web UI — React + Vite, dark-mode-first. Configure everything via point-and-click; nothing requires editing YAML. - One-command local install — docker compose up and the whole thing is running locally on http://localhost:3020. How it's built today Today's docker compose ships a deliberately compact two-container shape: one for the nginx-served React bundle, one for a FastAPI process that ALSO runs the in-process scheduler, the in-process compute, and a local SQLite file (persisted to a Docker volume) — the same single-binary pattern Plausible / Vaultwarden / n8n use for self-hosters. Great for laptops, fine for a small-team production deploy. The architecture is designed to split apart cleanly: every external dependency (storage, DB, secrets, auth, queue, compute) is meant to sit behind a Protocol with multiple implementations, so the same codebase can target separate api / worker / runtime containers backed by RDS / S3 / SecretsManager / SQS / Fargate when you outgrow the single-binary shape. Most of those Protocol extractions are still ahead of us — see PLATFORM_ARCHITECTURE.md for the gap matrix and the phasing plan. We're publishing the foundation now and growing it in the open. --- Quickstart — local install in 5 minutes Prerequisites: Docker Desktop 4.27+ (or Docker Engine 25.0+ with Compose v2.24+). That's it. bash git clone https://github.com/pycaret/pycaret.git cd pycaret docker compose up --build First build takes ~5 min (Python deps + npm install + Vite build). Subsequent starts are 10 min | Slow network on first build, or Docker Desktop is RAM-starved | Open Docker Desktop → Settings → Resources → bump RAM to ≥6 GB, retry | | no space left on device during build | Old Docker images filling the disk | docker system prune -a then retry | | /bin/sh^M: bad interpreter in api logs | Shell script got CRLF endings on a Windows checkout | This shouldn't happen — .gitattributes forces LF. If it does, re-clone with git clone --config core.autocrlf=false | | UI loads but every /api call returns 404 | pycaret-api container hasn't finished bootstrapping (Alembic migrations) — happens on the very first up | Wait ~30 sec for the health check to pass; docker compose logs -f api to wat
Top issues
#4216[Bug] `predict_model` silently drops `Anomaly_Score` for bare estimators because `decision_function` uses raw `X`1 comments · 2026-07-17#4223[BUG] tune_model fails for 'mlp' in RegressionExperiment with ValueError: Invalid parameter 'hidden_layer_size_2'0 comments · 2026-08-29#4190[Suggestion]: Bring back 'test_data' or add the ‘stratify’ parameter0 comments · 2026-05-14#4183[FEATURE]: Add privacy-aware business context for AI Assistant experiment design0 comments · 2026-05-03#4177PyCaret 4.0 is here — repository cleanup notice0 comments · 2026-04-27
Discussions
#4226[Security] Unauthenticated Localhost Fallback for MLflow Tracking URI0 comments · 2026-09-24#4225[Security] Remote Code Execution (RCE) via Unsafe Deserialization in load_model1 comments · 2026-09-24#4208Suggestion: Add PLS Regression (Partial Least Squares) to PyCaret0 comments · 2026-06-23#4193PyCaret returns slightly different results than scikit-learn2 comments · 2026-05-26#4167Performance metrics for binary and multiclass classification1 comments · 2025-06-08#4159Tuned model optimization for binary classification imbalanced data1 comments · 2025-03-31#4154PyCaret compare_models() Stuck at "Fitting 10 Folds"0 comments · 2025-02-25#4116pandas overlay during install0 comments · 2025-01-21
Latest release
PyCaret 3.3.2 3.3.2 2024-04-28Release: PyCaret 3.3.2 - Fixes installation and AUC metric (as well as other probability based metric) issues. What's Changed Update memory.py - removing joblib._format_load_msg() usage by @hakan-77 in https://github.com/pycaret/pycaret/pull/3980 Fix pycaret[full] cannot be installed by @jmakov in https://github.com/pycaret/pycaret/pull/3982 Fix AUC metric by @celestinoxp in https://github.com/pycaret/pycaret/pull/3935 New Contributors @hakan-77 made their first contribution in https://github.com/pycaret/pycaret/pull/3980 @jmakov made their first contribution in https://gi