keras-team/keras · error · ValueError

`label_weights` for multilabel data should be handled outsid

Error message

`label_weights` for multilabel data should be handled outside of `update_confusion_matrix_variables` when `multi_label` is True.

What it means

update_confusion_matrix_variables() is the shared engine behind Precision/Recall/confusion-matrix metrics. When multi_label=True it refuses a label_weights argument, because per-label weighting must be applied by the caller before state updates. Passing both raises this ValueError.

Source

Thrown at keras/src/metrics/metrics_utils.py:409

        `variables_to_update` must have a second dimension equal to the number
        of labels in y_true and y_pred, and those tensors must not be
        RaggedTensors.
      label_weights: (optional) tensor of non-negative weights for multilabel
        data. The weights are applied when calculating TP, FP, FN, and TN
        without explicit multilabel handling (i.e. when the data is to be
        flattened).
      thresholds_distributed_evenly: Boolean, whether the thresholds are evenly
        distributed within the list. An optimized method will be used if this is
        the case. See _update_confusion_matrix_variables_optimized() for more
        details.

    Raises:
      ValueError: If `y_pred` and `y_true` have mismatched shapes, or if
        `sample_weight` is not `None` and its shape doesn't match `y_pred`, or
        if `variables_to_update` contains invalid keys.
    """
    if multi_label and label_weights is not None:
        raise ValueError(
            "`label_weights` for multilabel data should be handled "
            "outside of `update_confusion_matrix_variables` when "
            "`multi_label` is True."
        )
    if variables_to_update is None:
        return
    if not any(
        key for key in variables_to_update if key in list(ConfusionMatrix)
    ):
        raise ValueError(
            "Please provide at least one valid confusion matrix "
            "variable to update. Valid variable key options are: "
            f'"{list(ConfusionMatrix)}". '
            f'Received: "{variables_to_update.keys()}"'
        )

    variable_dtype = list(variables_to_update.values())[0].dtype

View on GitHub (pinned to 7a34a03db6)

Solutions

1. Drop label_weights from the call and apply per-label weights yourself (multiply y_true or sample_weight per column before updating).
2. Alternatively pre-multiply sample_weight by the label weights and pass the result as sample_weight.

Example fix

# before
metrics_utils.update_confusion_matrix_variables(
    variables, y_true, y_pred, multi_label=True, label_weights=w)

# after
weighted = y_true * w  # apply per-label weights outside
metrics_utils.update_confusion_matrix_variables(
    variables, weighted, y_pred, multi_label=True)

def update_safe(**kw):
    if kw.get('multi_label') and kw.get('label_weights') is not None:
        kw.pop('label_weights')  # apply weights outside instead
    return metrics_utils.update_confusion_matrix_variables(**kw)

When it happens

Trigger: Calling update_confusion_matrix_variables(..., multi_label=True, label_weights=<array>) directly, e.g. from a custom multilabel metric that forwards both parameters.

Common situations: Writing a custom multilabel Precision/Recall variant and copying the single-label label_weights logic into the multilabel path.

Related errors

• Please provide at least one valid confusion matrix variable
• Invalid keys: "{invalid_keys}". Valid variable key options a
• ConvNeXt does not support the `channels_first` image data fo
• Invalid tensor type: {tensor_type}
• Layer `add_metric()` method is deprecated. Add your metric i

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/585d4f624ac3f952. Report an issue: GitHub.