PyCaret
GitHub
Resources

FAQs

Common questions about PyCaret 4.0.

Is PyCaret 4.0 backward-compatible with 3.x?#

No. The functional API (setup(), compare_models(), … as module-level calls) was removed. The OOP API (ClassificationExperiment etc.) replaces it entirely. See Migrate from 3.x for the full diff.

If you have existing 3.x code, the cleanest path is: pin to 3.x in that codebase and start fresh on 4.0 for new work. Mixing is not supported.

What Python versions does 4.0 support?#

3.11, 3.12, 3.13. We dropped 3.7-3.10 because they're either EOL or about to be. The codebase uses Python 3.11+ syntax (PEP 604 union types, native generics) without polyfills.

What sklearn version?#

≥ 1.7. The 4.0 engine uses APIs that landed in sklearn 1.6+ (root_mean_squared_error, the new set_output semantics, _USE_FLOAT32 internals).

Do I have to install the dashboard / backend?#

No. The core engine (pip install pycaret) is everything you need for notebook / script work. The dashboard + backend are opt-in via pip install "pycaret[dashboard]".

How do I do <thing-from-3.x-that-no-longer-exists>?#

Skim the What's removed section on each function page — Initialize, Train, Optimize, Analyze, Deploy, Other functions.

Most removed verbs have direct replacements (a sklearn API, a dedicated library, or a constructor parameter). If you can't find the replacement, open an issue and we'll either restore the feature or document the workaround.

Why no setup_kwargs escape hatch?#

3.x's setup() accepted ~100 keyword arguments. Many were undocumented, several were broken, and the long tail let users silently pass typos. 4.0 raises ConfigurationError on any unknown kwarg — explicit failures are better than silently-ignored config.

If you need a knob we don't have, file an issue. Adding it as a first-class constructor parameter is two lines of code.

Can I use my own preprocessing?#

Yes. Pass preprocess=False to skip the built-in chain entirely:

exp = ClassificationExperiment(target="y", preprocess=False).fit(already_clean_data)

You're then responsible for the train/test split's preprocessing parity — sklearn fit-on-train-then-transform-test discipline applies.

How do I add a custom model?#

Pass it to create_model directly:

from your_lib import YourClassifier
res = exp.create_model(YourClassifier(some_param=42))

The model is fitted, CV runs against it, and the resulting Pipeline is a real sklearn Pipeline. To make it visible to compare_models, register a container (see the source under pycaret/containers/models/); we'll write a more thorough plugin guide once the API stabilizes.

How do I add a custom metric?#

exp.add_metric(id, name, score_func). See Other functions for the full signature.

Where did evaluate_model go?#

Replaced by the dashboard's Model Card screen — same set of plots, but rendered in HTML/Plotly instead of an ipywidgets panel.

If you want a notebook-only equivalent, the individual functions under pycaret.plots.classification / pycaret.plots.regression return plotly.graph_objects.Figure objects you can .show().

Where did the gitbook docs go?#

Right here — you're reading them. The 3.x gitbook (pycaret.gitbook.io) is frozen and won't be updated. Everything from 4.0 onward lives in this repo's apps/site/ directory and is auto-deployed to pycaret.org on every push.

How do agents maintain the docs?#

See apps/site/AGENTS.md. Short version: drop MDX files in content/docs/<section>/, append session blocks to docs/revamp/release_notes_pycaret4.md, edit the repo-root CHANGELOG.md. Everything else is regenerated on each CI build.

Is there a paid version?#

No. PyCaret is MIT-licensed open source. There's no paid tier, no commercial license, no enterprise plan. Maintenance is funded by community contributions and sponsorship — not by gating features.

How do I cite PyCaret in a paper?#

@misc{pycaret,
  author       = {Moez Ali and contributors},
  title        = {PyCaret: An open-source, low-code machine learning library in Python},
  year         = {2024},
  howpublished = {\url{https://github.com/pycaret/pycaret}},
}

Where to ask for help?#