CategoryEncoding generate output with wrong shape when output_mode='one_hot' and the last dim is 1

[AGFACBNNR]

As shown in this colab, when output_mode='one_hot' and the last dim is 1, as described in Keras 3 Doc,

If the last dimension is size 1, will encode on that dimension.

The compute_output_shape() generate a correct static shape, while the output actually is generated by appending a new dimension for the encoded output, which is different from the documents.

Currently the output is:

I've tried it on tensorflow and jax backend and got the same result.
The results in line 1 of output shows the symptom I described above.

[SuryanarayanaY]

Hi @AGFACBNNR ,

I have replicated the issue and observed the discrepancy wrt actual output shape and shape from compute_output_shape method. Attached gist for reference.

Need to dig more and comeback. Thanks!