keras-team/keras · error · ValueError

The `num_bins` for `Hashing` cannot be `None` or non-positiv

Error message

The `num_bins` for `Hashing` cannot be `None` or non-positive values. Received: num_bins={num_bins}.

What it means

Hashing maps values into num_bins buckets, so num_bins must be a positive integer. None, zero, or negative values make the hash space empty and are rejected at construction.

Source

Thrown at keras/src/layers/preprocessing/hashing.py:161

        sparse=False,
        **kwargs,
    ):
        if not tf.available:
            raise ImportError(
                "Layer Hashing requires TensorFlow. "
                "Install it via `pip install tensorflow`."
            )

        # By default, output int32 when output_mode='int' and floats otherwise.
        if "dtype" not in kwargs or kwargs["dtype"] is None:
            kwargs["dtype"] = (
                "int64" if output_mode == "int" else backend.floatx()
            )

        super().__init__(**kwargs)

        if num_bins is None or num_bins <= 0:
            raise ValueError(
                "The `num_bins` for `Hashing` cannot be `None` or "
                f"non-positive values. Received: num_bins={num_bins}."
            )

        if output_mode == "int" and (
            self.dtype_policy.name not in ("int32", "int64")
        ):
            raise ValueError(
                'When `output_mode="int"`, `dtype` should be an integer '
                f"type, 'int32' or 'in64'. Received: dtype={kwargs['dtype']}"
            )

        # 'output_mode' must be one of (INT, ONE_HOT, MULTI_HOT, COUNT)
        accepted_output_modes = ("int", "one_hot", "multi_hot", "count")
        if output_mode not in accepted_output_modes:
            raise ValueError(
                "Invalid value for argument `output_mode`. "
                f"Expected one of {accepted_output_modes}. "

View on GitHub (pinned to 7a34a03db6)

Solutions

Set num_bins to a positive integer larger than expected cardinality (common practice: a power of two like 2**18)
Compute num_bins defensively: num_bins = max(1, estimated_cardinality)
Do not leave num_bins unset in custom wrappers that forward None

Example fix

// before
layer = Hashing(num_bins=len(vocab) - len(vocab))  # evaluates to 0
// after
layer = Hashing(num_bins=max(2**18, len(vocab)))

Defensive patterns

Strategy: validation

Validate before calling

assert isinstance(num_bins, int) and num_bins > 0, "num_bins must be a positive int"

Type guard

def valid_num_bins(n):
    return isinstance(n, int) and n > 0

Try / catch

catch ValueError from Hashing.__init__ and correct num_bins before constructing again

Prevention

Set num_bins to a positive integer (e.g. 2**18 for high-cardinality hashing)
Size num_bins above expected cardinality to limit collisions

When it happens

Trigger: Hashing(num_bins=0), Hashing(num_bins=-1), or Hashing(num_bins=None) — the constructor validates num_bins is a positive number.

Common situations: Passing num_bins=0 by mistake; deriving num_bins from data as len(set(x)) when the set is empty; config typo like num_bins=-1 or forgetting the argument (None).

Understand the failure class

Background: "Must be a positive integer", "Invalid value", "Unsupported": the invalid-argument-value error family, when a library rejects the value you pass — this error's family across 35 libraries.

Related errors

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/9dcdc4d74d6520a7. Report an issue: GitHub.