Fail to load model for iemocap evaluation. about delta HOT 4 CLOSED

see this tensorflow/tensorflow#30892.
I will check if eval can work, train_and_eval is recommended.
we using TF keras not Keras.

from delta.

I don't think the issue is related. It seems rather to have something to do with dynamic shapes. I found an (awful) hacky workaround by adding

self.model.fit({'inputs': np.zeros((1,700,257,1)), 'labels':np.array([0])}, epochs=0)

in asr_solver.build() after self.model.compile(). This seems to cause all layers of the model to be build properly, while compile leaves something unbuilt.

from delta.

relevant issue keras-team/keras#11072

from delta.

This error is cause by .h5 model , please using tensorflow model instead.
Fix model save from callback and model load, replacing *.h5 by *.ckpt suffix.

from delta.