Tim Head betatim

AGENTS Instruction

This project is called foobar. Its goal is to provide ...

This file contains is additional guidance for AI agents and other AI editors.

Core Principles

Array API Architecture

Created: 2026-01-07 Last Updated: 2026-01-07

Overview

Scikit-learn's Array API support enables estimators and functions to work with arrays from different libraries (NumPy, CuPy, PyTorch) without modification. This allows computations to run on GPUs when using GPU-backed array libraries.

The implementation follows the Array API Standard, a specification that defines a common API for array manipulation libraries.

GitHub PR Activity Tracker

A Python script that tracks Pull Request activity for a specific user over a configurable time period using the PyGithub library.

Features

Tracks PRs where the user:
- Created the PR
- Added comments
Submitted reviews

AI-Assisted PR Review Checklist for Scikit-learn

Purpose: This checklist is optimized for AI assistants (like Cursor) to perform automated PR reviews. It separates automatable checks from those requiring human judgment, provides specific patterns to detect, and includes commands to run.

How to Use This Checklist

For AI Agents:

Run all AUTOMATED checks first and report findings with severity levels

Summary of Issues

Classification Metrics Sparse Support Bug (Issue #32036): A bug where classification metrics in scikit-learn claim sparse matrix support in docstrings but raise an error when used with sparse inputs. The issue is reliably reproducible with provided code steps, expected (support) vs. actual behavior (TypeError), and environment details in the traceback. No major missing elements. Link
RandomizedSearchCV Feature Request (Issue #32032): A proposal to add weights for controlling the probability of selecting items in a list of parameter distributions, useful for complex pipelines with interdependent hyperparameters. This is a feature enhancement, not a bug, and includes clear examples and rationale. Link
CI Failure on Linux Build (Issue #32022): Reported CI failure on a specific build configuration, with a reference to logs but no detailed steps to rep

	"""
	Test: Validate that cuml native estimators can be converted to ONNX
	via as_sklearn() -> skl2onnx -> onnxruntime.

	Unlike cuml.accel proxies (which skl2onnx recognizes directly), native cuml
	estimators must first be converted to sklearn via as_sklearn() before
	skl2onnx.convert_sklearn() will accept them.

	Run without cuml.accel:
	python test_onnx_as_sklearn.py

	#!/usr/bin/env python3
	"""
	Ray + RandomForestClassifier with max_calls=1

	Demonstrates the impact of max_calls=1 on Ray task execution when using
	scikit-learn's RandomForestClassifier.
	"""
	import time
	import ray
	from sklearn.datasets import make_classification

	"""
	Benchmark: scikit-learn RandomForest vs LightGBM RandomForest

	Compares performance across:
	- Number of samples (1K, 10K, 100K, 500K)
	- Number of features (10, 50, 200)
	- Feature types (numerical, categorical, mixed)
	- Number of classes (2, 5, 10)

	Includes cases optimized for LightGBM's strengths:

	name: tabareana-20251202
	channels:
	- conda-forge
	dependencies:
	- _libgcc_mutex=0.1=conda_forge
	- _openmp_mutex=4.5=2_gnu
	- bzip2=1.0.8=hda65f42_8
	- ca-certificates=2025.11.12=hbd8a1cb_0
	- cuda-cccl_linux-64=13.0.85=ha770c72_0
	- cuda-cudart-dev_linux-64=13.0.96=h376f20c_0

	from __future__ import annotations

	import warnings
	warnings.simplefilter("error", FutureWarning)
	from pathlib import Path
	from typing import Any

	import pandas as pd

	from tabarena.benchmark.experiment import AGModelBagExperiment, ExperimentBatchRunner