Tim Head betatim

name	pr-impact-brief
description	Assemble source-linked evidence on whether a scikit-learn PR is worth carrying, one PR at a time, ending in the questions the decision hinges on and never in a verdict. Use when the user says "let's review some PRs for worth", asks to review or triage PRs for worth or impact, asks whether a PR is worth it or worth merging or worth carrying, or asks for an impact brief on a PR number or URL.

PR Impact Brief

Walk one unattended scikit-learn PR at a time and present the evidence a maintainer needs to answer "should we carry this?" — then stop. The human decides.

This skill exists because the cost side of a change is invisible by default. A PR that is small, correct, and green still adds permanent public surface, consumes review time, and displaces bigger work. Nobody asks whether it is worth that, so the question goes unanswered for years.

name	pr-walkthrough
description	Produce a structured walkthrough of a GitHub PR ordered by causal dependency. Use when the user asks to walk through, analyze, or understand a PR's changes.

PR Walkthrough

Generate a causal-dependency-ordered walkthrough of a GitHub PR and write it to agents/reviews/ in the repository.

This skill is scoped to PRs in the current repository. Cross-repo PRs are out of scope.

name	pr-walkthrough
description	Produce a structured walkthrough of a GitHub PR ordered by causal dependency. Use when the user asks to walk through, analyze, or understand a PR's changes.

PR Walkthrough

Generate a causal-dependency-ordered walkthrough of a GitHub PR and write it to agents/reviews/ in the repository.

This skill is scoped to PRs in the current repository. Cross-repo PRs are out of scope.

AGENTS Instruction

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

Core Principles

These principles reduce common LLM coding mistakes. Apply them to every task.

	# SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
	# SPDX-License-Identifier: Apache-2.0
	"""Summarise a SkillEvaluator tier3 run from its per-trial scores.

	SkillEvaluator only reports a score for a condition (with or without the skill)
	when every trial in it succeeded. One agent timeout or provider error leaves the
	report at N/A and no ``lift.json`` is written, even though the other trials were
	graded. This script reads each trial's ``reward.json`` instead. It skips trials
	that errored or have no reward, and says how many were skipped.

	"""Sweep pipeline `fit` and `predict` runtime against the size of X.

	This script times

	StandardScaler -> Nystroem -> LogisticRegression

	as a function of how much data it is given, in two independent sweeps:

	fit GridSearchCV(pipeline, PARAM_GRID, cv=3).fit(X[:n], y[:n]),
	for n from 2_000 to 100_000

	numpy
	as-is as numpy difference added
	search 1.18s 1.18s -0.0% -164us
	predict 24.2ms 19.1ms -20.9% -5.1ms
	serve/call 897us 930us +3.7% 33us
	score 10.4ms 10.9ms +4.8% 493us
	one 10-row predict, step by step:
	scale.transform 111us 135us +22.2% 24us
	kernel.transform 572us 588us +2.8% 16us
	clf.predict 206us 202us -2.2% -5us

	"""Benchmark storing fitted attributes as NumPy.

	Times a PCA + LogisticRegression pipeline with and without the
	`store_fitted_as_numpy` configuration option added in
	https://github.com/scikit-learn/scikit-learn/pull/34667, for
	https://github.com/scikit-learn/scikit-learn/issues/34604.

	Needs that branch checked out. Run it with no arguments:

	python bench_store_fitted_as_numpy.py

	import numpy as np
	from sklearn.ensemble import RandomForestRegressor
	from sklearn.model_selection import cross_validate
	from skrub import tabular_pipeline
	from skrub.datasets import fetch_employee_salaries


	def main():
	print("Loading employee salaries dataset ...")
	data = fetch_employee_salaries()

	#!/usr/bin/env python3
	"""
	Employee Salary Prediction
	==========================
	Predict current annual salary for Montgomery County employees
	using a scikit-learn pipeline with mixed feature types.
	"""

	try:
	import cuml.accel