Train, save model and load: error while loading model

Question

I'm training a model using TensorFlow and Keras. I would like to save the model and then load it. but I'm getting some errors.

I compile the model this way:

from tensorflow.keras.models import Model
import tensorflow as tf

model.compile(loss='categorical_crossentropy',
              optimizer=adam,
              metrics=['accuracy', top3, top5])

and after training I'm loading the model this way:

model.save('model')

So I get a folder "model" containing:

---model 
     ---assets
     ---variables
     ---keras_metadata.pb
     ---saved_model.pb

Finally, I try to load the model using:

import tensorflow as tf

new_model = tf.keras.models.load_model('model')
new_model.summary()

But I'm getting this error:

ValueError: Unable to restore custom object of type _tf_keras_metric 
currently. Please make sure that the layer implements `get_config`and 
`from_config` when saving. In addition, please use the `custom_objects` 
arg when calling `load_model()`.

I tried that too. I got: OSError: Unable to open file (bad object header version number) when doing: new_model = tf.keras.models.load_model('model.h5') — Laura
– Laura, Commented Jun 9, 2021 at 21:24

afsharov · Accepted Answer · 2021-06-09 22:51:53Z

1

When your model utilizes custom objects, such as your custom metrics, you have to specify them with the custom_objects argument of load_model:

new_model = tf.keras.models.load_model('model', custom_objects={'top3': top3, 'top5': top5})

Note that the definitions of your custom metrics must be available in the same module/environment you are loading the model.

edited Jun 9, 2021 at 22:51

answered Jun 9, 2021 at 22:44

afsharov

5,1842 gold badges14 silver badges29 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

afsharov Over a year ago

@Laura did the answer resolve your problem?

Collectives™ on Stack Overflow

Train, save model and load: error while loading model

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related