Dataset Viewer
Auto-converted to Parquet Duplicate
task_id
stringclasses
18 values
sample_id
stringlengths
42
63
source_step
int64
1
274
action_type
stringclasses
2 values
messages
listlengths
2
2
aerial-cactus-identification__00
aerial-cactus-identification__00__sample_00001
1
code
[ { "role": "user", "content": "# Introduction\n\nYou are a Kaggle grandmaster attending a competition. In order to win this competition, you need to come up with an excellent and creative plan for a solution and then implement this solution in Python. We will now provide a description of the task.\n\n# Task ...
aerial-cactus-identification__00
aerial-cactus-identification__00__sample_00002
3
code
[ { "role": "user", "content": "# Introduction\n\nYou are a Kaggle grandmaster attending a competition. In order to win this competition, you need to come up with an excellent and creative plan for a solution and then implement this solution in Python. We will now provide a description of the task.\n\n# Task ...
aerial-cactus-identification__00
aerial-cactus-identification__00__sample_00003
4
code
[ { "role": "user", "content": "# Introduction\n\nYou are a Kaggle grandmaster attending a competition. In order to win this competition, you need to come up with an excellent and creative plan for a solution and then implement this solution in Python. We will now provide a description of the task.\n\n# Task ...
aerial-cactus-identification__00
aerial-cactus-identification__00__sample_00004
6
code
[{"role":"user","content":"# Introduction\n\nYou are a Kaggle grandmaster attending a competition. Y(...TRUNCATED)
aerial-cactus-identification__00
aerial-cactus-identification__00__sample_00005
7
code
[{"role":"user","content":"# Introduction\n\nYou are a Kaggle grandmaster attending a competition. Y(...TRUNCATED)
aerial-cactus-identification__00
aerial-cactus-identification__00__sample_00006
8
code
[{"role":"user","content":"# Introduction\n\nYou are a Kaggle grandmaster attending a competition. Y(...TRUNCATED)
aerial-cactus-identification__00
aerial-cactus-identification__00__sample_00007
9
code
[{"role":"user","content":"# Introduction\n\nYou are a Kaggle grandmaster attending a competition. Y(...TRUNCATED)
aerial-cactus-identification__00
aerial-cactus-identification__00__sample_00008
10
code
[{"role":"user","content":"# Introduction\n\nYou are a Kaggle grandmaster attending a competition. Y(...TRUNCATED)
aerial-cactus-identification__00
aerial-cactus-identification__00__sample_00009
12
code
[{"role":"user","content":"# Introduction\n\nYou are a Kaggle grandmaster attending a competition. Y(...TRUNCATED)
aerial-cactus-identification__00
aerial-cactus-identification__00__sample_00010
13
code
[{"role":"user","content":"# Introduction\n\nYou are a Kaggle grandmaster attending a competition. Y(...TRUNCATED)
End of preview. Expand in Data Studio

长程机器学习任务轨迹样本

训练数据

本数据集包含 807 条训练样本、149 条验证样本,来自 10 个源竞赛家族的 22 个构造任务。每条样本由解题过程中可见的历史上下文和当前教师响应组成,覆盖代码生成与执行后 review。

数据由 DeepSeek v4.1 Flash 通过 AIDE 采集,开启思考模式,每任务最长 24 小时、最多 500 个探索执行步骤。样本保留真实思考、代码、原生 review 及当时可见的执行反馈,适用于长上下文推理和代码代理的监督微调。

加载与字段

from datasets import load_dataset

dataset = load_dataset("jankin123/mle-data-long-1106")
字段 含义
task_id 任务标识,对应任务说明、配置和数据包
sample_id 唯一的匿名样本编号
source_step 执行轨迹中的动作步号
action_type code 或 review
messages 两条消息:历史 user 上下文、当前 assistant 响应
messages[].loss user 为 false,assistant 为 true

训练集包含 374 个 code、433 个 review;验证集包含 70 个 code、79 个 review。同一步可能同时包含代码生成和执行后 review,不能按 task_id + source_step 合并样本。

assistant 内容保留 <think>...</think>,随后是完整 Python 程序或原生 submit_review 工具调用文本。仅当前 assistant 响应参与监督,历史上下文不参与当前响应的损失计算。使用时按目标模型的 tokenizer 和模板进行编码。

任务与划分

源竞赛家族 训练任务 验证任务 训练样本 验证样本
aerial-cactus-identification 1 0 42 0
detecting-insults-in-social-commentary 3 0 125 0
dog-breed-identification 1 0 35 0
jigsaw-toxic-comment-classification-challenge 3 1 126 19
nomad2018-predict-transparent-conductors 2 0 92 0
plant-pathology-2020-fgvc7 1 1 52 66
spooky-author-identification 2 1 90 37
tabular-playground-series-dec-2021 2 0 89 0
tabular-playground-series-may-2022 1 1 30 27
text-normalization-challenge-english-language 2 0 126 0
合计 18 4 807 149

划分按完整任务固定,同一任务的样本保持在同一侧。同来源的不同变体可能共享源行,因此不是按源竞赛家族或全部源行隔离的划分。

防污染与脱敏

题目仅从指定的 MLE-bench 公开 TRAIN 池构造,排除原基准 test/private 数据及其标签。单题内部按文本内容、图像内容、表格特征或句子分组,防止相应重复内容跨越训练/隐藏分区。解题阶段只使用公开输入、执行日志、训练集内部验证和提交格式反馈,隐藏评分在候选冻结后独立进行。

发布版清理身份标识、凭证、部署路径、网络地址及图像/归档元数据,并保持任务输入与评分标识的对应关系。未完成语义近重复过滤、外部题库交叉去污染及模型预训练污染检测。处理范围与局限见 数据处理说明。

构造数据的环境

environment/ 提供这批数据对应的题目构造环境,从源 TRAIN 池生成训练输入、无目标测试输入、提交样例和隐藏评分答案。构造本身只需要 CPU,无需模型 API。

train.jsonl
validation.jsonl
tasks.jsonl                         # 任务索引、划分与样本统计
tasks/<task_id>/task.md              # 题目说明
tasks/<task_id>/task.json            # 输入、目标和评分契约
task_data/<task_id>.tar.gz           # 已发布的任务数据包
environment/
  README.md                         # 安装、数据准备与使用方法
  build_tasks.py                    # 从源 TRAIN 重新构造题目
  prepare_task.py                    # 解包已发布题目
  grade_constructed.py              # 使用明确任务配置独立评分
  construction/                     # 分组、划分及文件构造
  configs/construction.json         # 来源、字段、种子和任务配置
  requirements/construction.txt
  Dockerfile
  docs/DATA_PROCESSING.md

使用已发布的题目

python environment/prepare_task.py \
  --task detecting-insults-in-social-commentary__00 --output task_work

解题程序读取 task_work/input/,将提交写入 task_work/submission/submission.csv;grading/ 应由评分端保管,不向解题程序提供。独立评分示例:

python environment/grade_constructed.py \
  --task-json tasks/detecting-insults-in-social-commentary__00/task.json \
  --task-dir task_work \
  --predictions task_work/submission/submission.csv

从源数据构造题目

python -m pip install -r environment/requirements/construction.txt

python environment/build_tasks.py --all --plan

python environment/build_tasks.py \
  --task detecting-insults-in-social-commentary__00 \
  --source-root /path/to/source-data \
  --output /path/to/new-task-build --archive

源数据应位于 <source-root>/<competition-id>/prepared/public/。构造输出保留源 TRAIN 原生值,已发布任务包则经过脱敏;两者的目标与划分对应,但部分标识和文件名不同。

完整步骤见 题目构造环境。源数据的使用条件沿用相应竞赛规定。

Downloads last month
194