constraint_dimension_metrics_test.json

{
  "family": "router",
  "split": "test",
  "overall": {
    "rows": 6000,
    "accuracy": 0.8828333333333334,
    "macro_f1": 0.883185453547077,
    "weighted_f1": 0.883185453547077,
    "micro_f1": 0.8828333333333334,
    "labels": [
      "constraint_dimension::causal",
      "constraint_dimension::goal",
      "constraint_dimension::other",
      "constraint_dimension::policy",
      "constraint_dimension::state",
      "constraint_dimension::value"
    ],
    "confusion_matrix": [
      [
        966,
        2,
        20,
        1,
        7,
        4
      ],
      [
        1,
        927,
        47,
        2,
        6,
        17
      ],
      [
        12,
        30,
        817,
        22,
        56,
        63
      ],
      [
        0,
        2,
        26,
        949,
        1,
        22
      ],
      [
        27,
        7,
        63,
        8,
        865,
        30
      ],
      [
        7,
        19,
        137,
        27,
        37,
        773
      ]
    ],
    "calibration_error": 0.04340881209580012,
    "classification_report": {
      "constraint_dimension::causal": {
        "precision": 0.9536031589338598,
        "recall": 0.966,
        "f1-score": 0.9597615499254843,
        "support": 1000.0
      },
      "constraint_dimension::goal": {
        "precision": 0.939209726443769,
        "recall": 0.927,
        "f1-score": 0.9330649219929542,
        "support": 1000.0
      },
      "constraint_dimension::other": {
        "precision": 0.7360360360360361,
        "recall": 0.817,
        "f1-score": 0.7744075829383886,
        "support": 1000.0
      },
      "constraint_dimension::policy": {
        "precision": 0.9405351833498513,
        "recall": 0.949,
        "f1-score": 0.944748631159781,
        "support": 1000.0
      },
      "constraint_dimension::state": {
        "precision": 0.8899176954732511,
        "recall": 0.865,
        "f1-score": 0.8772819472616633,
        "support": 1000.0
      },
      "constraint_dimension::value": {
        "precision": 0.8503850385038504,
        "recall": 0.773,
        "f1-score": 0.8098480880041907,
        "support": 1000.0
      },
      "accuracy": 0.8828333333333334,
      "macro avg": {
        "precision": 0.8849478064567696,
        "recall": 0.8828333333333332,
        "f1-score": 0.883185453547077,
        "support": 6000.0
      },
      "weighted avg": {
        "precision": 0.8849478064567696,
        "recall": 0.8828333333333334,
        "f1-score": 0.883185453547077,
        "support": 6000.0
      }
    }
  },
  "per_task": {
    "constraint_dimension": {
      "rows": 6000,
      "accuracy": 0.8828333333333334,
      "macro_f1": 0.883185453547077,
      "weighted_f1": 0.883185453547077,
      "micro_f1": 0.8828333333333334,
      "labels": [
        "causal",
        "goal",
        "other",
        "policy",
        "state",
        "value"
      ],
      "confusion_matrix": [
        [
          966,
          2,
          20,
          1,
          7,
          4
        ],
        [
          1,
          927,
          47,
          2,
          6,
          17
        ],
        [
          12,
          30,
          817,
          22,
          56,
          63
        ],
        [
          0,
          2,
          26,
          949,
          1,
          22
        ],
        [
          27,
          7,
          63,
          8,
          865,
          30
        ],
        [
          7,
          19,
          137,
          27,
          37,
          773
        ]
      ],
      "wrong_task_predictions": 0,
      "wrong_task_rate": 0.0
    }
  },
  "calibration": {
    "method": "temperature_grid_search",
    "rows": 6000,
    "temperature": 2.8,
    "loss": null
  }
}