competition-data-science
Prepare datasets and operate RD-Agent data-science/Kaggle workflows with reproducible debug, validation, submission, and grading artifacts.
- License 许可
- MIT
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- competition-data-science
- license
- MIT
- metadata
- { "disco-role": "operating", "parent-skill": "rd-agent" }
- description
- Prepare datasets and operate RD-Agent data-science/Kaggle workflows with reproducible debug, validation, submission, and grading artifacts.
- disable-model-invocation
- true
02 资源文件 1
- 参考文档references/dataset-contract.md1.2 KB
03 SKILL.md 正文
RD-Agent competition and data science
Use this sub-skill for rdagent data_science, Kaggle-style tasks, custom datasets, data preparation, generated modeling code, local evaluation, and submission artifacts.
Input contract
A custom task should be organized around a task directory and a source_data root. The repository's example documents the following shape:
source_data/<task>/prepare.py
<task>/<task>/description.md
<task>/<task>/sample.py # optional debug sampler
<task>/eval/grade.py # optional test grader
<task>/eval/valid.py # optional submission validator
The task description should state the objective, data fields, split/use restrictions, modeling constraints, metric, and submission format. Run check_dataset_layout.py before asking an agent to write modeling code.
Safe workflow
- Inspect
rdagent data_science --helpand the scenario config. - Run
prepare.pyor the dataset preparation step in a disposable output location; verify row counts, columns, labels, and train/test separation. - Create a tiny debug sample using the task's
sample.pywhen present. The default sampler inrdagent.scenarios.data_science.debug.datais a fallback, not a guarantee of semantic correctness. - Launch the smallest agent loop that exercises code generation and evaluation.
- Validate the generated submission with
valid.pywhen available, then grade withgrade.pyor the configured evaluator. - Preserve the description, resolved paths, generated source, submission, validation output, metric, and seed.
Competition hygiene
- Do not use test labels, competition metadata, or future information in feature engineering.
- Check that local validation matches the competition's target and that a generated submission has the exact required columns/order.
- Treat external Kaggle credentials and downloads as prerequisites; do not include them in a reusable skill or test fixture.
- A successful code-generation iteration is not a score. Report the evaluator command and the data split that produced the score.
- For time-series or patient data, preserve group/entity boundaries and temporal ordering while sampling.
Read dataset-contract.md for required files and artifact semantics. For Qlib-backed finance tasks use quant-finance instead.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.