name: DataSelectEnv
version: "1.0.0"
description: >
  OpenEnv RL environment for data curation in ML training.
  Agents learn to select high-quality training data from a noisy pool
  under budget constraints, balancing uncertainty, diversity, and noise
  avoidance to incrementally improve a classifier's validation performance.

tags:
  - openenv
  - active-learning
  - data-curation
  - noisy-labels
  - machine-learning
  - reinforcement-learning

authors:
  - InfraNova

observation_space:
  type: object
  properties:
    remaining_budget:
      type: integer
      description: Samples remaining in the selection budget
    diversity_score:
      type: number
      description: Standard deviation of the current training set (proxy for diversity)
    noise_estimate:
      type: number
      description: Estimated fraction of noisy samples remaining in the pool
    current_performance:
      type: number
      description: Current model validation performance (1 / (1 + log_loss))
    samples_available:
      type: integer
      description: Number of samples remaining in the unlabeled pool

action_space:
  type: object
  required:
    - action_type
    - batch_size
    - strategy_weights
  properties:
    action_type:
      type: string
      enum: [select_batch, stop]
      description: Select a batch of data or stop the episode early
    batch_size:
      type: integer
      minimum: 0
      description: Number of samples to select this step
    strategy_weights:
      type: object
      description: Weights for each sampling strategy (normalized internally)
      properties:
        uncertainty:
          type: number
          minimum: 0.0
        diversity:
          type: number
          minimum: 0.0
        random:
          type: number
          minimum: 0.0

tasks:
  - id: easy
    difficulty: easy
    description: >
      Clean dataset (flip_y=0.05), budget=300, max_steps=15.
      Agent must reach validation performance > 0.62.
      Score is normalized over range [0.55, 0.75].
    success_criteria: "current_performance > 0.62"

  - id: medium
    difficulty: medium
    description: >
      High noise (flip_y=0.25), budget=150, max_steps=12.
      Agent must reach performance > 0.52 while keeping average
      noise selection rate below 0.50.
    success_criteria: "current_performance > 0.52 AND avg_noise_ratio < 0.50"

  - id: hard
    difficulty: hard
    description: >
      High noise (flip_y=0.30), tight budget=100, max_steps=8.
      Agent must hit performance > 0.58 efficiently.
      Grader scores performance and budget efficiency jointly.
    success_criteria: "current_performance > 0.58, scored jointly with efficiency"

reward:
  type: continuous
  range: [-inf, +inf]
  description: >
    Shaped reward combining performance gain, diversity bonus,
    redundancy penalty, noise penalty, and budget cost.
    Provides dense signal throughout the episode — not just at termination.

endpoints:
  websocket: WS   /ws       # primary transport; required on HF Spaces
  reset:   POST /reset
  step:    POST /step
  state:   GET  /state
  tasks:   GET  /tasks
  grader:  POST /grader
  baseline: GET /baseline
  health:  GET  /health