keras-team/keras · error · ValueError

Expected the two input tensors to have identical shapes. Rec

Error message

Expected the two input tensors to have identical shapes. Received input_shape={input_shape}

What it means

HashedCrossing requires its two input tensors to have identical shapes (the last dimension must match). Mismatched shapes make the crossing ill-defined, so compute_output_shape rejects them.

Source

Thrown at keras/src/layers/preprocessing/hashed_crossing.py:121

        self.num_bins = num_bins
        self.output_mode = output_mode
        self.sparse = sparse
        self._allow_non_tensor_positional_args = True
        self._convert_input_args = False
        self.supports_jit = False

    def compute_output_shape(self, input_shape):
        if (
            not len(input_shape) == 2
            or not isinstance(input_shape[0], tuple)
            or not isinstance(input_shape[1], tuple)
        ):
            raise ValueError(
                "Expected as input a list/tuple of 2 tensors. "
                f"Received input_shape={input_shape}"
            )
        if input_shape[0][-1] != input_shape[1][-1]:
            raise ValueError(
                "Expected the two input tensors to have identical shapes. "
                f"Received input_shape={input_shape}"
            )

        if not input_shape:
            if self.output_mode == "int":
                return ()
            return (self.num_bins,)
        if self.output_mode == "int":
            return tuple(input_shape[0])

        if self.output_mode == "one_hot" and input_shape[0][-1] != 1:
            return tuple(input_shape[0]) + (self.num_bins,)

        return tuple(input_shape[0])[:-1] + (self.num_bins,)

    def call(self, inputs):
        from keras.src.backend import tensorflow as tf_backend

View on GitHub (pinned to 7a34a03db6)

Solutions

Make both inputs the same shape, typically (batch, 1)
Reshape one input upstream: tf.reshape(x, [-1, 1]) or a keras Reshape((1,)) layer
Check Input(shape=...) declarations in Functional models for equality

Example fix

// before
a = keras.Input(shape=(1,)); b = keras.Input(shape=(2,))
out = HashedCrossing(100)([a, b])
// after
a = keras.Input(shape=(1,)); b = keras.Input(shape=(1,))
out = HashedCrossing(100)([a, b])

Defensive patterns

Strategy: validation

Validate before calling

assert input_shape[0][-1] == input_shape[1][-1], "inputs must have identical last dim"

Type guard

def same_last_dim(s):
    return s[0][-1] == s[1][-1]

Try / catch

catch ValueError from build/compute_output_shape and reshape both inputs to a common shape before retrying

Prevention

Give both inputs to HashedCrossing identical shapes, e.g. (batch, 1) each
Reshape mismatched inputs to (batch, 1) with tf.reshape before feeding the layer

When it happens

Trigger: input_shape[0][-1] != input_shape[1][-1], e.g. Input(shape=(1,)) and Input(shape=(2,)) both wired into the crossing layer.

Common situations: Two Input layers with different shapes (e.g. (1,) and (2,)); one input reshaped upstream and the other not.

Related errors

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/f57a4d7843fe2550. Report an issue: GitHub.