PyCaret

Warnings

• breaking PyCaret 3.0 introduced significant API changes, making code written for PyCaret 2.x largely incompatible without modifications. [4, 9]
  Fix: Refer to the PyCaret 3.x documentation and tutorials for updated API usage. Consider rewriting or adapting existing PyCaret 2.x code.
• breaking Python 3.7 support was dropped in PyCaret 3.1.0, and Python 3.8 support was dropped in PyCaret 3.3.0. [4]
  Fix: Ensure your Python environment is running Python 3.9 or newer.
• breaking The `deep_check` and `eda` functions were removed in PyCaret 3.1.0 and will raise exceptions if called. [4]
  Fix: Remove calls to `deep_check()` and `eda()`. Use alternative data exploration or preprocessing methods if needed.
• gotcha The `setup()` function is interactive by default, prompting the user for data type confirmation. This can halt execution in non-interactive environments. [1, 10]
  Fix: For automated scripts, pass `silent=True` to suppress prompts. Always include `session_id` for reproducibility, e.g., `setup(data=data, target='target_col', session_id=123, silent=True)`.
• gotcha PyCaret 3.x has had specific dependency version constraints. For instance, PyCaret 3.0.4 pinned `scikit-learn<1.3.0` and 3.0.1 pinned `numpy<1.24`. While newer PyCaret 3.x versions (e.g., 3.3.0) support `scikit-learn 1.4` and `pandas 2.0`, users might encounter conflicts if not using the latest compatible PyCaret version. [4, 9]
  Fix: Always install PyCaret in a clean virtual environment. Consult the official documentation for specific dependency requirements for your PyCaret version, or upgrade to the latest PyCaret to benefit from broader dependency support.
• gotcha The behavior and results of `compare_models()` might have changed significantly between PyCaret 3.0.0 and versions >= 3.0.1 due to internal bug fixes affecting model performance evaluation. [18]
  Fix: If upgrading from 3.0.0, re-evaluate model performance and selection as results may differ. The changes were intended to be bug fixes, leading to more accurate evaluations in later versions.

Install

• pip install pycaret Basic Installation
• pip install pycaret[full] Full Installation (includes all optional dependencies)

Imports

• get_data

  from pycaret.datasets import get_data

• setup

  from pycaret.classification import setup

  PyCaret's functional API requires importing specific modules (e.g., `classification`, `regression`) to access functions like `setup`.
• ClassificationExperiment

  from pycaret.classification import ClassificationExperiment

  For the object-oriented API introduced in PyCaret 3.x.
• compare_models

  from pycaret.classification import compare_models

• predict_model

  from pycaret.classification import predict_model

• save_model

  from pycaret.classification import save_model

Quickstart

This quickstart demonstrates a typical PyCaret classification workflow using the functional API. It involves loading a dataset, initializing the environment with `setup()`, comparing multiple models with `compare_models()`, making predictions, and saving the best performing model. [1, 7, 10, 11]

import pandas as pd
from pycaret.datasets import get_data
from pycaret.classification import *

# Load a sample dataset
data = get_data('diabetes')

# Initialize the setup (classification experiment)
# Use silent=True for non-interactive environments and session_id for reproducibility
clf1 = setup(data=data, target='Class variable', session_id=123, silent=True)

# Compare all available models and select the best one
best_model = compare_models()

# Make predictions on the hold-out set
predictions = predict_model(best_model)

# Save the trained model pipeline
save_model(best_model, 'diabetes_best_pipeline')