Dhrona1421/multimodal-content-moderation

1"""2tasks.py — Task registry and factory for the Content Moderation Environment.3"""4from __future__ import annotations5from typing import Any, Dict6from env import ContentModerationEnv7 8TASKS: Dict[str, Dict[str, Any]] = {9    "easy": {10        "name": "Easy Moderation",11        "description": (12            "Obvious cases only. All modality signals align. "13            "Strong keyword-matching baselines score well here."14        ),15        "difficulty":              "easy",16        "max_steps":               12,17        "reward_range":            [0.0001, 0.9999],18        "dataset_subset":          "12 sampled from 14 easy posts",19        "expected_baseline_score": 0.88,20    },21    "medium": {22        "name": "Intermediate Moderation",23        "description": (24            "Contextual reasoning required. Coded hate speech, "25            "health misinformation, trust-level signals matter."26        ),27        "difficulty":              "medium",28        "max_steps":               12,29        "reward_range":            [0.0001, 0.9999],30        "dataset_subset":          "12 sampled from 13 medium posts",31        "expected_baseline_score": 0.68,32    },33    "hard": {34        "name": "Expert Moderation",35        "description": (36            "Adversarial edge cases with conflicting signals: "37            "safe text + harmful image, trusted users spreading misinfo, "38            "suspicious users with innocent content."39        ),40        "difficulty":              "hard",41        "max_steps":               12,42        "reward_range":            [0.0001, 0.9999],43        "dataset_subset":          "12 sampled from 14 hard posts",44        "expected_baseline_score": 0.52,45    },46}47 48 49def make_task(50    task_name:    str,51    dataset_path: str = "moderation_dataset.json",52    seed:         int = 42,53) -> ContentModerationEnv:54    if task_name not in TASKS:55        raise ValueError(f"Unknown task '{task_name}'. Available: {list(TASKS.keys())}")56    cfg = TASKS[task_name]57    return ContentModerationEnv(58        dataset_path=dataset_path,59        task=cfg["difficulty"],60        max_steps=cfg["max_steps"],61        seed=seed,62    )63 64 65def list_tasks() -> Dict[str, Dict[str, Any]]:66    return {name: dict(cfg) for name, cfg in TASKS.items()}67 68 69def describe_task(task_name: str) -> str:70    if task_name not in TASKS:71        raise ValueError(f"Unknown task '{task_name}'.")72    cfg = TASKS[task_name]73    return (74        f"Task        : {cfg['name']}\n"75        f"Difficulty  : {cfg['difficulty'].upper()}\n"76        f"Max Steps   : {cfg['max_steps']}\n"77        f"Dataset     : {cfg['dataset_subset']}\n"78        f"Reward      : {cfg['reward_range']}\n"79        f"Description : {cfg['description']}\n"80    )81 82 83if __name__ == "__main__":84    for name in TASKS:85        print(describe_task(name))86        print()87