PREM BABU KANAPARTHI
Update: fix username links to PremC1F and premxai/emotion_engine
6619824 verified | """ | |
| emotion_engine.py — HuggingFace Datasets loading script | |
| Registered as the dataset builder for the emotion-engine HF repo. | |
| Supports loading by condition name or "all" for the full dataset. | |
| Usage: | |
| from datasets import load_dataset | |
| # Single condition | |
| ds = load_dataset("PremC1F/emotion-engine", "condition_d") | |
| # All conditions concatenated | |
| ds = load_dataset("PremC1F/emotion-engine", "all") | |
| """ | |
| import datasets | |
| _CITATION = """ | |
| @article{kanaparthi2026emotion, | |
| author = {Kanaparthi, Prem Babu}, | |
| title = {Emergent Emotional Appraisal in Generative Agents | |
| via Predictive World Modeling}, | |
| year = {2026}, | |
| } | |
| """ | |
| _DESCRIPTION = """ | |
| 204,520 agent-step records from a five-condition controlled experiment comparing | |
| emotion-engine architectures in a 12-agent ghost-survival simulation (Ghost Town). | |
| Five conditions: | |
| baseline_0 — no emotion system | |
| condition_a — hand-coded OCC appraisal rules | |
| condition_b — behavioral cloning from condition_a | |
| condition_c — emotion dynamics model | |
| condition_d — predictive world modeling (proposed) | |
| Six scenarios per condition: | |
| standard_night, high_ghost_pressure, storm_scarcity, | |
| ally_death, betrayal_refusal, crowded_shelter | |
| 8 random seeds per scenario. | |
| Each record contains: full 14-dim observation, 6-dim emotion vector, | |
| 8-dim latent representation, 12 future-event predictions (Condition D), | |
| action taken, and outcome metadata. | |
| """ | |
| _HOMEPAGE = "https://github.com/PremC1F/emotion-engine" | |
| _LICENSE = "MIT" | |
| _CONDITIONS = [ | |
| "baseline_0", | |
| "condition_a", | |
| "condition_b", | |
| "condition_c", | |
| "condition_d", | |
| ] | |
| _FEATURES = datasets.Features({ | |
| # Identity | |
| "run_id": datasets.Value("string"), | |
| "step": datasets.Value("int32"), | |
| "agent_id": datasets.Value("string"), | |
| "condition": datasets.Value("string"), | |
| "seed": datasets.Value("int32"), | |
| "scenario": datasets.Value("string"), | |
| # Observation (14 dims) | |
| "obs_visible_ghosts": datasets.Value("int32"), | |
| "obs_visible_deaths": datasets.Value("int32"), | |
| "obs_nearby_allies": datasets.Value("int32"), | |
| "obs_trusted_allies": datasets.Value("int32"), | |
| "obs_supplies_seen": datasets.Value("int32"), | |
| "obs_in_shelter": datasets.Value("bool"), | |
| "obs_nearest_refuge_distance": datasets.Value("float32"), | |
| "obs_steps_since_ghost_seen": datasets.Value("int32"), | |
| "obs_steps_since_ally_died": datasets.Value("int32"), | |
| "obs_steps_since_betrayal": datasets.Value("int32"), | |
| "obs_ally_deaths_witnessed": datasets.Value("int32"), | |
| "obs_betrayals_received": datasets.Value("int32"), | |
| "obs_average_trust": datasets.Value("float32"), | |
| "obs_graph_tension": datasets.Value("float32"), | |
| # Affect / emotion | |
| "affect_fear": datasets.Value("float32"), | |
| "affect_grief": datasets.Value("float32"), | |
| "affect_trust": datasets.Value("float32"), | |
| "affect_stress": datasets.Value("float32"), | |
| "affect_relief": datasets.Value("float32"), | |
| "affect_suspicion": datasets.Value("float32"), | |
| # Latent representation (8 dims) | |
| "latent_0": datasets.Value("float32"), | |
| "latent_1": datasets.Value("float32"), | |
| "latent_2": datasets.Value("float32"), | |
| "latent_3": datasets.Value("float32"), | |
| "latent_4": datasets.Value("float32"), | |
| "latent_5": datasets.Value("float32"), | |
| "latent_6": datasets.Value("float32"), | |
| "latent_7": datasets.Value("float32"), | |
| # Future-event predictions (Condition D; zero for others) | |
| "pred_ghost_nearby_t3": datasets.Value("float32"), | |
| "pred_my_death_t5": datasets.Value("float32"), | |
| "pred_health_drop_t3": datasets.Value("float32"), | |
| "pred_nearby_death_t5": datasets.Value("float32"), | |
| "pred_help_success_t5": datasets.Value("float32"), | |
| "pred_refusal_received_t5": datasets.Value("float32"), | |
| "pred_tie_increase_t5": datasets.Value("float32"), | |
| "pred_shelter_achieved_t2": datasets.Value("float32"), | |
| "pred_storm_onset_t3": datasets.Value("float32"), | |
| "pred_scarcity_t5": datasets.Value("float32"), | |
| "pred_graph_tension_t3": datasets.Value("float32"), | |
| "pred_valence_t5": datasets.Value("float32"), | |
| # Action & outcome | |
| "action": datasets.Value("string"), | |
| "goal": datasets.Value("string"), | |
| "reward_survival": datasets.Value("float32"), | |
| "reward_shelter": datasets.Value("float32"), | |
| "reward_health": datasets.Value("float32"), | |
| "reward_social": datasets.Value("float32"), | |
| "total_reward": datasets.Value("float32"), | |
| "alive": datasets.Value("bool"), | |
| "health": datasets.Value("float32"), | |
| "sheltered": datasets.Value("bool"), | |
| "time_of_day": datasets.Value("string"), | |
| }) | |
| class EmotionEngineConfig(datasets.BuilderConfig): | |
| def __init__(self, condition="all", **kwargs): | |
| super().__init__(**kwargs) | |
| self.condition = condition | |
| class EmotionEngine(datasets.GeneratorBasedBuilder): | |
| """Ghost Town emotion engine dataset — 204,520 agent-step records.""" | |
| VERSION = datasets.Version("1.0.0") | |
| BUILDER_CONFIG_CLASS = EmotionEngineConfig | |
| BUILDER_CONFIGS = [ | |
| EmotionEngineConfig( | |
| name="all", | |
| version=VERSION, | |
| description="All five conditions concatenated (204,520 records)", | |
| condition="all", | |
| ), | |
| ] + [ | |
| EmotionEngineConfig( | |
| name=c, | |
| version=VERSION, | |
| description=f"Condition {c} only (~40k records, 6 scenarios, 8 seeds)", | |
| condition=c, | |
| ) | |
| for c in _CONDITIONS | |
| ] | |
| DEFAULT_CONFIG_NAME = "condition_d" | |
| def _info(self): | |
| return datasets.DatasetInfo( | |
| description=_DESCRIPTION, | |
| features=_FEATURES, | |
| homepage=_HOMEPAGE, | |
| license=_LICENSE, | |
| citation=_CITATION, | |
| ) | |
| def _split_generators(self, dl_manager): | |
| if self.config.condition == "all": | |
| files = { | |
| c: dl_manager.download(f"data/{c}.parquet") | |
| for c in _CONDITIONS | |
| } | |
| return [ | |
| datasets.SplitGenerator( | |
| name=datasets.Split.TRAIN, | |
| gen_kwargs={"filepaths": list(files.values())}, | |
| ) | |
| ] | |
| else: | |
| filepath = dl_manager.download( | |
| f"data/{self.config.condition}.parquet" | |
| ) | |
| return [ | |
| datasets.SplitGenerator( | |
| name=datasets.Split.TRAIN, | |
| gen_kwargs={"filepaths": [filepath]}, | |
| ) | |
| ] | |
| def _generate_examples(self, filepaths): | |
| import pandas as pd | |
| idx = 0 | |
| for path in filepaths: | |
| df = pd.read_parquet(path) | |
| for _, row in df.iterrows(): | |
| yield idx, row.to_dict() | |
| idx += 1 | |