dataclean-env / openenv.yaml
Dhruv Goyal
DataClean OpenEnv v2.0 - Data Drift + Parallel Baseline
782a588
Raw
History Blame Contribute Delete
1.4 kB
name: dataclean-env
version: 2.0.0
description: >
Real-world data cleaning and ETL environment for AI agent RL training.
4 tasks including the novel Data Drift task — live streaming rows injected
mid-episode every 5 steps, simulating a Kafka pipeline under continuous drift.
author: Team DataClean
tags:
- openenv
- data-cleaning
- etl
- real-world
- tabular
- data-drift
- streaming
tasks:
- id: task1
name: Null Fixer
difficulty: easy
max_steps: 10
description: Fix null values and wrong dtypes in a 50-row customer dataset.
- id: task2
name: Schema Normalizer
difficulty: medium
max_steps: 20
description: Deduplicate, normalize country names, fix date format, fill nulls.
- id: task3
name: ETL Pipeline
difficulty: hard
max_steps: 30
description: Merge two tables, remove outliers, add derived column.
- id: task4_data_drift
name: Data Drift (Streaming)
difficulty: expert
max_steps: 40
description: >
NOVEL. Live streaming transactions — fresh dirty rows injected every 5 steps
mid-episode. Agent must clean continuously under real-time data drift.
environment:
type: http
port: 7860
endpoints:
reset: POST /reset
step: POST /step
state: GET /state
tasks: GET /tasks
grader: GET /grader
baseline: GET /baseline
health: GET /health