keras-team/keras · error · ValueError

Invalid AUC curve value: "{key}". Expected values are ["PR",

Error message

Invalid AUC curve value: "{key}". Expected values are ["PR", "ROC", "PRGAIN"]

What it means

AUC accepts its curve parameter as an AUCCurve enum or a case-insensitive string. AUCCurve.from_str() recognizes only the keys roc/ROC, pr/PR and prgain/PRGAIN, and raises this ValueError for anything else. Names like 'precision-recall' or 'ROCC' are rejected.

Source

Thrown at keras/src/metrics/metrics_utils.py:58

class AUCCurve(Enum):
    """Type of AUC Curve (ROC or PR)."""

    ROC = "ROC"
    PR = "PR"
    PRGAIN = "PRGAIN"

    @staticmethod
    def from_str(key):
        if key in ("pr", "PR"):
            return AUCCurve.PR
        elif key in ("roc", "ROC"):
            return AUCCurve.ROC
        elif key in ("prgain", "PRGAIN"):
            return AUCCurve.PRGAIN
        else:
            raise ValueError(
                f'Invalid AUC curve value: "{key}". '
                'Expected values are ["PR", "ROC", "PRGAIN"]'
            )


class AUCSummationMethod(Enum):
    """Type of AUC summation method.

    https://en.wikipedia.org/wiki/Riemann_sum)

    Contains the following values:
    * 'interpolation': Applies mid-point summation scheme for `ROC` curve. For
      `PR` curve, interpolates (true/false) positives but not the ratio that is
      precision (see Davis & Goadrich 2006 for details).
    * 'minoring': Applies left summation for increasing intervals and right
      summation for decreasing intervals.
    * 'majoring': Applies right summation for increasing intervals and left
      summation for decreasing intervals.

View on GitHub (pinned to 7a34a03db6)

Solutions

1. Use exactly 'ROC', 'PR', or 'PRGAIN' (case-insensitive).
2. For a precision-recall AUC pass curve='PR'; for gain curves pass 'PRGAIN'.
3. Or pass the enum directly: keras.metrics.AUC(curve=AUCCurve.PR) (the string form is preferred).

Example fix

# before
auc = keras.metrics.AUC(curve='precision-recall')

# after
auc = keras.metrics.AUC(curve='PR')

VALID_CURVES = {'roc', 'pr', 'prgain'}
def check_curve(c):
    if isinstance(c, str) and c.lower() not in VALID_CURVES:
        raise ValueError(f'curve must be one of {sorted(VALID_CURVES)}')
    return c

def is_valid_auc_curve(c) -> bool:
    return not isinstance(c, str) or c.lower() in {'roc', 'pr', 'prgain'}

When it happens

Trigger: Calling keras.metrics.AUC(curve='precision-recall') or AUC(curve='sketch-roc') - any string other than the roc/pr/prgain variants.

Common situations: Assuming scikit-learn style naming ('precision', 'roc_auc_score') transfers to Keras; typo'd values coming from config files.

Understand the failure class

Background: Invalid enum value errors: "Unknown type", "Invalid scope", "must be one of" — when a string is not on the library's allowed list — this error's family across 23 libraries.

Related errors

• Invalid AUC summation method value: "{key}". Expected values
• Invalid `curve` argument value "{curve}". Expected one of: {
• Invalid `summation_method` argument value "{summation_method
• Argument `num_thresholds` must be an integer > 1. Received:
• `num_labels` is needed only when `multi_label` is True.

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/4d2ca336bf2a512f. Report an issue: GitHub.