Installation

PyCaret 4.0 supports Python 3.11, 3.12, and 3.13. The core engine installs in seconds and pulls in only thirteen direct dependencies (down from sixty-five in 3.x).

From PyPI#

pip install pycaret

This installs the engine — every classification, regression, clustering, anomaly, and time-series API. To run the optional dashboard or backend, install the matching extras:

pip install "pycaret[dashboard]"   # FastAPI server + worker + DB
pip install "pycaret[explain]"     # SHAP for advanced explainability
pip install "pycaret[forecast]"    # Extra sktime adapters for TS work

From source#

git clone https://github.com/pycaret/pycaret.git
cd pycaret
uv sync             # or: pip install -e packages/engine[dev]
uv run pytest packages/engine/tests

We use uv for development. It is not required to use PyCaret — only to develop it.

Verify the install#

from pycaret.classification import ClassificationExperiment
from pycaret.datasets import get_data

data = get_data("juice", verbose=False)
exp = ClassificationExperiment(target="Purchase").fit(data)
print(exp.create_model("lr").metrics)

If that prints a five-row metrics DataFrame, you're ready. Continue to the quickstart for the full workflow.

Hardware notes#

• CPU: PyCaret runs everywhere scikit-learn does. No special hardware required.
• GPU: Set use_gpu=True on any Experiment to opt into GPU-accelerated estimators (XGBoost / LightGBM / cuML when installed). The default path stays on CPU.
• Memory: 4 GB is comfortable for tutorial-scale data; 16 GB+ is what you'd want for serious work.

Upgrading from 3.x#

PyCaret 4.0 is a clean break: the functional API (setup() / compare_models() / etc. as module-level calls) was removed. The OOP API (ClassificationExperiment, RegressionExperiment, …) replaces it entirely. See Migrate from 3.x for the full diff.