results/episode_task_suite/temporal_order/metrics.json

{
  "accuracy": 0.4540229885057471,
  "precision": 0.4665271966527197,
  "recall": 0.6408045977011494,
  "f1": 0.5399515738498789,
  "tp": 223,
  "tn": 93,
  "fp": 255,
  "fn": 125,
  "positive_rate_true": 0.5,
  "positive_rate_pred": 0.6867816091954023,
  "task": "temporal_order",
  "input": "two adjacent windows -> whether order is correct",
  "split": "chronological",
  "num_samples": 2320,
  "num_train_samples": 1624,
  "num_test_samples": 696,
  "train_final_accuracy": 0.5086206896551724,
  "task_display_name": "Temporal Order Verification"
}