datalore-clean

分类: 数据与AI | 上传者: micic-mihajlo | 下载: 0 | 版本: v1.0（最新）

当任务需要在本地数据集上进行确定性数据清理或导出时使用，例如填充缺失的数值、删除空行、去重、选择列、重命名列、派生列、过滤行、排序、标准化或写入清理的 CSV。

更新日志: Source: GitHub https://github.com/micic-mihajlo/Datalore

目录结构

当前层级: tree/main/.agents/skills/datalore-clean/

📄 SKILL.md 1.2 KB

SKILL.md

---
name: datalore-clean
description: Use when a task needs deterministic data cleaning or export on a local dataset, such as filling missing numeric values, dropping null rows, deduplicating, selecting columns, renaming columns, deriving columns, filtering rows, sorting, standardizing, or writing a cleaned CSV.
---

Prefer the Datalore clean command over custom pandas for routine transformations.

Preferred command:
```bash
./datalore clean <dataset-path> --format json
```

Useful flags:
- `--fill-numeric mean|median|zero`
- `--drop-na`
- `--drop-duplicates`
- `--select-columns col_a,col_b,col_c`
- `--rename old=>new`
- `--derive new_col=>left|op|right`
- `--filter 'customers>=24'`
- `--sort-by column`
- `--descending`
- `--limit 100`
- `--standardize`
- `--output artifacts/datasets/<name>.csv`
- `--report-file artifacts/reports/<name>.json`

Rules:
1. Use the CLI result as the source of truth for what changed.
2. Prefer explicit `--output` and `--report-file` when later steps need stable paths.
3. When a later command depends on the cleaned file, run the steps sequentially, not in parallel.
4. Only write custom pandas when the requested transformation is outside the supported deterministic flags.

登录后下载/点赞/收藏 ❤ 42 | ★ 0

datalore-clean

目录结构

SKILL.md

举报内容

提示