name: data-cleaning-openenv
version: 1.0.0
description: >
  An OpenEnv-compliant environment where AI agents learn to clean
  messy real-world datasets step by step. Tasks include duplicate
  removal, missing value imputation, dtype fixing, outlier detection,
  and schema validation.

author: thorodin103
license: mit

tags:
  - openenv
  - data-cleaning
  - real-world
  - pandas
  - tabular
  - reinforcement-learning
  - agent

api:
  base_url: https://thorodin103-data-cleaning-openenv.hf.space
  reset: POST /reset/{task_id}
  step: POST /step/{task_id}
  state: GET /state/{task_id}
  tasks: GET /tasks
  validate: GET /validate
  health: GET /health

tasks:
  - id: easy_dedup_rename
    difficulty: easy
    description: Remove duplicate rows and rename columns to snake_case
    max_steps: 10
    scoring:
      duplicate_score: 0.5
      schema_score: 0.5

  - id: medium_missing_dtype
    difficulty: medium
    description: Fill missing values and fix wrong data types
    max_steps: 15
    scoring:
      missing_score: 0.5
      dtype_score: 0.5

  - id: hard_full_pipeline
    difficulty: hard
    description: Full pipeline — duplicates, missing, dtypes, outliers, schema
    max_steps: 20
    scoring:
      duplicate_score: 0.2
      missing_score: 0.2
      dtype_score: 0.2
      outlier_score: 0.2
      schema_score: 0.2

observation_space:
  task_id: string
  step: integer
  dataset_info: object
  columns: list
  shape: list
  missing_values: object
  dtypes: object
  duplicate_count: integer
  sample_rows: list
  available_operations: list
  task_description: string
  message: string

action_space:
  operation:
    type: string
    values:
      - remove_duplicates
      - fill_missing
      - fix_dtype
      - remove_outliers
      - rename_columns
      - validate_schema
      - finish
  parameters:
    type: object

reward_space:
  total:
    type: float
    range: [0.0, 1.0]
  components:
    - duplicate_score
    - missing_score
    - dtype_score
    - outlier_score
    - schema_score
    - penalty

baseline_scores:
  model: gpt-4o-mini
  easy_dedup_rename: 1.0000
  medium_missing_dtype: 0.6643
  hard_full_pipeline: 0.8386
  average: 0.8343

runtime:
  language: python
  version: "3.10"
  framework: fastapi
  port: 7860