[BUG] Conflicting `loss_weights` implementation in Keras3 for single output case. about keras

[fchollet]

Thanks for the report. This should be fixed in the latest Keras release (3.3.0), please check.

[fchollet]

Do you have a code snippet to reproduce the issue?

[proneetsharma]

To reproduce the issue, you can run the training script available here. Despite using a custom model and metrics in this scenario, the problem persists even when employing Keras inherited metrics such as keras.metrics.BinaryAccuracy().

For clarity, Keras inherited metrics were utilized in the provided description. The issue does not seem to be related to custom loss metrics, as Keras 3 expects a function in the form of result = fn(y_true, y_pred), which our metric fulfills. It appears to be specifically related to the type of metrics used for the single output case.

The script requires two additional installations:

• pip install pypaz --user
• pip install fiftyone

Steps to run:

1. python download_openimagesV6.py
2. python train.py

The training script works fine with keras2.

[proneetsharma]

Thank you @fchollet !

[SuryanarayanaY]

Hi @proneetsharma ,

Please feel free to close the issue if resolved. Thanks!

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No