constraint_dimension_metrics_eval.json

{
  "family": "router",
  "split": "eval",
  "overall": {
    "rows": 6000,
    "accuracy": 0.8698333333333333,
    "macro_f1": 0.8703105353987516,
    "weighted_f1": 0.8703105353987517,
    "micro_f1": 0.8698333333333333,
    "labels": [
      "constraint_dimension::causal",
      "constraint_dimension::goal",
      "constraint_dimension::other",
      "constraint_dimension::policy",
      "constraint_dimension::state",
      "constraint_dimension::value"
    ],
    "confusion_matrix": [
      [
        965,
        2,
        16,
        0,
        12,
        5
      ],
      [
        1,
        907,
        57,
        1,
        9,
        25
      ],
      [
        19,
        31,
        791,
        18,
        56,
        85
      ],
      [
        1,
        0,
        22,
        954,
        2,
        21
      ],
      [
        27,
        13,
        78,
        6,
        842,
        34
      ],
      [
        13,
        23,
        145,
        23,
        36,
        760
      ]
    ],
    "calibration_error": 0.04364704496164792,
    "classification_report": {
      "constraint_dimension::causal": {
        "precision": 0.9405458089668616,
        "recall": 0.965,
        "f1-score": 0.9526159921026653,
        "support": 1000.0
      },
      "constraint_dimension::goal": {
        "precision": 0.9293032786885246,
        "recall": 0.907,
        "f1-score": 0.9180161943319838,
        "support": 1000.0
      },
      "constraint_dimension::other": {
        "precision": 0.7132551848512173,
        "recall": 0.791,
        "f1-score": 0.7501185395922239,
        "support": 1000.0
      },
      "constraint_dimension::policy": {
        "precision": 0.9520958083832335,
        "recall": 0.954,
        "f1-score": 0.9530469530469531,
        "support": 1000.0
      },
      "constraint_dimension::state": {
        "precision": 0.8798328108672936,
        "recall": 0.842,
        "f1-score": 0.8605007664793051,
        "support": 1000.0
      },
      "constraint_dimension::value": {
        "precision": 0.8172043010752689,
        "recall": 0.76,
        "f1-score": 0.7875647668393783,
        "support": 1000.0
      },
      "accuracy": 0.8698333333333333,
      "macro avg": {
        "precision": 0.8720395321387332,
        "recall": 0.8698333333333332,
        "f1-score": 0.8703105353987516,
        "support": 6000.0
      },
      "weighted avg": {
        "precision": 0.8720395321387332,
        "recall": 0.8698333333333333,
        "f1-score": 0.8703105353987517,
        "support": 6000.0
      }
    }
  },
  "per_task": {
    "constraint_dimension": {
      "rows": 6000,
      "accuracy": 0.8698333333333333,
      "macro_f1": 0.8703105353987516,
      "weighted_f1": 0.8703105353987517,
      "micro_f1": 0.8698333333333333,
      "labels": [
        "causal",
        "goal",
        "other",
        "policy",
        "state",
        "value"
      ],
      "confusion_matrix": [
        [
          965,
          2,
          16,
          0,
          12,
          5
        ],
        [
          1,
          907,
          57,
          1,
          9,
          25
        ],
        [
          19,
          31,
          791,
          18,
          56,
          85
        ],
        [
          1,
          0,
          22,
          954,
          2,
          21
        ],
        [
          27,
          13,
          78,
          6,
          842,
          34
        ],
        [
          13,
          23,
          145,
          23,
          36,
          760
        ]
      ],
      "wrong_task_predictions": 0,
      "wrong_task_rate": 0.0
    }
  },
  "calibration": {
    "method": "temperature_grid_search",
    "rows": 6000,
    "temperature": 2.8,
    "loss": null
  }
}