---
title: "TraceML：人类与智能体在机器学习开发中规划行为的实证分析"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.26086v1"
permalink: "https://feed.astralor.com/items/itm_a363e546de7c6fe0491347a6"
published: "2026-08-26T17:50:13.000Z"
score: 76
channel: "ai"
tags: ["Kaggle","MLEvolve","OpenAI Codex","TraceML","智能体规划","机器学习开发"]
---

# TraceML：人类与智能体在机器学习开发中规划行为的实证分析

## AI 点评

> 这篇论文最值得看的地方是把‘差距’从分数变成了行为轨迹，让我们能直观看到Agent在长时间任务中如何陷入固定的循环。对于做Agent推理、规划或AutoML的工程师，这份开源数据集可能比一大堆基准更有肌肉记忆价值。建议配合人类轨迹一起看，很多提示词设计会变得更有依据。

## 速览

大型语言模型在独立编程问题上已能写出正确代码，但在自主机器学习开发中仍远逊于人类。现有基准只评估最终提交，忽略了发展过程。TraceML 引入了统一的版本级模式，涵盖 134 个 Kaggle 竞赛中 4,465 条人类轨迹，以及两个智能体在七个竞赛中的 430 条配对人类轨迹和 207 条智能体轨迹。每个代码版本都带有分数、时间戳及操作标签。分析发现，人类会灵活切换数据、验证、模型和集成策略，而智能体陷入狭窄循环。基于人类实践提炼的简短规划提示可部分弥合差距，但无法完全复现人类行为。

## 关键点

- TraceML 数据集包含 4,465 条人类 Kaggle 轨迹和 207 条智能体轨迹，覆盖 134 个竞赛，所有版本均标注分数、时间戳、操作与意图。
- 实证发现人类专家在数据工作、验证、模型更改和集成之间灵活切换，并会重新打开之前搁置的方案，而两个智能体 scaffold 均陷入狭窄循环。
- 从人类实践中提炼的简短规划提示能提升智能体分数，但无法改变其整体行为轮廓，说明指令只弥合了可归因于指令的部分差距。
- 研究团队公开了语料库、模式、标注器和提取流程，便于后续研究复现与扩展。

## 全文

Large language models write correct code for isolated problems but remain far weaker at autonomous machine-learning development, where an agent must revise data pipelines, models, and validation over hours of feedback, and on most competitions still finishes below strong human competitors. Outcome-based benchmarks record this gap but not its cause, because they grade the final submission and discard the development process behind it. We introduce TraceML, which pairs human and agent work on the same competitions under one version-level schema: 4,465 human Kaggle trajectories across 134 competitions, seven of which are also worked by two agent scaffolds, giving 430 paired human and 207 agent trajectories. Every code version carries its score, its timestamp, and labels for the action taken, its intent, the edit size, and the score effect. Read this way, the gap becomes concrete. Experts alternate data work, validation, model changes, and ensembling, and return to approaches they had set aside. Each agent scaffold instead collapses into a narrow loop: Codex spends its steps re-weighting ensembles and tuning submissions, MLEvolve mutates its model in place, and neither pivots at the human rate nor reopens abandoned work. A short planning prompt distilled from human practice moves the behaviors it names toward the human profile and lifts scores, but the effort profile stays agent-shaped: instruction closes only the part of the gap that reduces to instructions. We release the corpus, the schema, the labelers, and the extraction pipeline at https://huggingface.co/datasets/jerryyan/TraceML.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_a363e546de7c6fe0491347a6