Closed caseyjconger closed 4 years ago
@dizcology PTAL
@caseyjconger for Hyper parameter tuning we use the TensorBoard callback. https://www.tensorflow.org/api_docs/python/tf/keras/callbacks/TensorBoard what happens is that this callback returns a summary name of epoch_ because of this: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/keras/callbacks.py#L2084 You may need to create custom callbacks and it summary operation to use other names.
I think there may be an issue with cloudml-samples/census/tf-keras/hptuning_config.yaml. It seems like the target metric is set to be
epoch_acc
butepoch_acc
isn't defined anywhere in the code. Should this just be 'accuracy' as that's the metric defined in the model?