confidence_bin_metrics_test.json

{
  "family": "router",
  "split": "test",
  "overall": {
    "rows": 3000,
    "accuracy": 1.0,
    "macro_f1": 1.0,
    "weighted_f1": 1.0,
    "micro_f1": 1.0,
    "labels": [
      "low",
      "medium",
      "high"
    ],
    "confusion_matrix": [
      [
        1000,
        0,
        0
      ],
      [
        0,
        1000,
        0
      ],
      [
        0,
        0,
        1000
      ]
    ],
    "calibration_error": 0.0,
    "classification_report": {
      "confidence_bin::high": {
        "precision": 1.0,
        "recall": 1.0,
        "f1-score": 1.0,
        "support": 1000.0
      },
      "confidence_bin::low": {
        "precision": 1.0,
        "recall": 1.0,
        "f1-score": 1.0,
        "support": 1000.0
      },
      "confidence_bin::medium": {
        "precision": 1.0,
        "recall": 1.0,
        "f1-score": 1.0,
        "support": 1000.0
      },
      "accuracy": 1.0,
      "macro avg": {
        "precision": 1.0,
        "recall": 1.0,
        "f1-score": 1.0,
        "support": 3000.0
      },
      "weighted avg": {
        "precision": 1.0,
        "recall": 1.0,
        "f1-score": 1.0,
        "support": 3000.0
      }
    },
    "ordinal_mae": 0.0,
    "off_by_two_rate": 0.0
  },
  "per_task": {
    "confidence_bin": {
      "rows": 3000,
      "accuracy": 1.0,
      "macro_f1": 1.0,
      "weighted_f1": 1.0,
      "micro_f1": 1.0,
      "labels": [
        "high",
        "low",
        "medium"
      ],
      "confusion_matrix": [
        [
          1000,
          0,
          0
        ],
        [
          0,
          1000,
          0
        ],
        [
          0,
          0,
          1000
        ]
      ],
      "wrong_task_predictions": 0,
      "wrong_task_rate": 0.0
    }
  }
}