keras-team/keras · error · ValueError

Invalid value for argument `num_regressors`. Expected a valu

Error message

Invalid value for argument `num_regressors`. Expected a value >= 0. Received: num_regressors={num_regressors}

What it means

R2Score requires num_regressors (the predictor count used for adjusted R-squared) to be a non-negative integer or None. Passing a negative value raises this ValueError in __init__. Pass None to get plain, non-adjusted R-squared.

Source

Thrown at keras/src/metrics/regression_metrics.py:427

        dtype=None,
    ):
        super().__init__(name=name, dtype=dtype)
        # Metric should be maximized during optimization.
        self._direction = "up"

        valid_class_aggregation_values = (
            None,
            "uniform_average",
            "variance_weighted_average",
        )
        if class_aggregation not in valid_class_aggregation_values:
            raise ValueError(
                "Invalid value for argument `class_aggregation`. Expected "
                f"one of {valid_class_aggregation_values}. "
                f"Received: class_aggregation={class_aggregation}"
            )
        if num_regressors < 0:
            raise ValueError(
                "Invalid value for argument `num_regressors`. "
                "Expected a value >= 0. "
                f"Received: num_regressors={num_regressors}"
            )
        self.class_aggregation = class_aggregation
        self.num_regressors = num_regressors
        self.num_samples = self.add_variable(
            shape=(),
            initializer=initializers.Zeros(),
            name="num_samples",
        )
        self._built = False

    def _build(self, y_true_shape, y_pred_shape):
        if len(y_pred_shape) != 2 or len(y_true_shape) != 2:
            raise ValueError(
                "R2Score expects 2D inputs with shape "
                "(batch_size, output_dim). Received input "

View on GitHub (pinned to 7a34a03db6)

Solutions

1. Pass num_regressors=None if you do not need adjusted R-squared.
2. Otherwise pass the actual input feature count, e.g. num_regressors=X.shape[1].
3. Map config sentinels like -1 to None before constructing the metric.

Example fix

# before
metric = keras.metrics.R2Score(num_regressors=-1)

# after
metric = keras.metrics.R2Score(num_regressors=X_train.shape[1])
# or keras.metrics.R2Score() for plain R2

if num_regressors is not None:
    assert isinstance(num_regressors, int) and num_regressors >= 0, 'num_regressors must be >= 0 or None'

def is_valid_num_regressors(v) -> bool:
    return v is None or (isinstance(v, int) and v >= 0)

When it happens

Trigger: keras.metrics.R2Score(num_regressors=-1) or any negative number, often from a config default of -1 meaning 'unset'.

Common situations: Config systems that use -1 as a sentinel for 'not configured'; arithmetic computing num_regressors from dataset dimensions that underflows.

Related errors

• Invalid value for argument `class_aggregation`. Expected one
• R2Score expects 2D inputs with shape (batch_size, output_dim
• R2Score expects 2D inputs with shape (batch_size, output_dim
• Layer `add_metric()` method is deprecated. Add your metric i
• Argument `num_thresholds` must be an integer > 0. Received:

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/c5b11b328e178fc5. Report an issue: GitHub.