Getting array of 'nan' after prediction on milestone project 1(food vision big).

[Osprey-DS]

hi, after doing the milestone project 1 and saving the model when I try to predict by loading it, it gives an array of 'nan' of 101 elements. can anyone help me where I have done the mistake? although i follow the exact same process as from course video and GitHub notebook,

array([[nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan]], dtype=float32)

[mrdbourke]

Hey Osprey,

I'd check the inputs going into your model, are they the right inputs?

Also, I'd step through the code to see where it starts to output different values.

Does the model still evaluate correctly?

For example, model.evaluate(test_data)?

Does the model work correctly before saving?

Is there anything else you've tried so far?

[Osprey-DS]

Yes daniels, the model evaluates correctly before saving or in the time when I build the model that time it gives correct. accuracy score a, as expected from your GitHub notebook and video,(around 80%), but when I saved the model and then load it through tf.keras.models.load_model it gives that output. (I saved it on hdf5 because of the web app.)
i am sharing the notebook here, you can check, and find the mistake.
Notebook
https://colab.research.google.com/drive/1_Ea8LOJxHNByCRl4ZZTX4QEontZV9-ls?usp=sharing

Here it gives nan values Notebook
https://colab.research.google.com/drive/11MJ0W6vfhuIVd4cQSf27lukTWMW6zEB_?usp=sharing

Although I will build the model again just by copy from the course GitHub notebook of yours and then I will let you know what happens this time.
Thank you..

[Osprey-DS]

I have created a new notebook and build the model this time the model works fine inside google colab,
the nan problem is solved here is the notebook https://colab.research.google.com/drive/15sJJhrZBo12CA3flnrX-NC4WwrP84z0D?usp=sharing

but the model is giving an error when I try to load it on my local machine, it is giving me optimizer error on local machine

Please Daniel help me with this. how I can solve this error.

tf.__version__
2.3.0

model = tf.keras.models.load_model('model_0_80.h5')

`---------------------------------------------------------------------------
KeyError Traceback (most recent call last)
in
----> 1 model = tf.keras.models.load_model('model_0_80.h5')

E:\TF_env\lib\site-packages\tensorflow\python\keras\saving\save.py in load_model(filepath, custom_objects, compile, options)
180 if (h5py is not None and (
181 isinstance(filepath, h5py.File) or h5py.is_hdf5(filepath))):
--> 182 return hdf5_format.load_model_from_hdf5(filepath, custom_objects, compile)
183
184 filepath = path_to_string(filepath)

E:\TF_env\lib\site-packages\tensorflow\python\keras\saving\hdf5_format.py in load_model_from_hdf5(filepath, custom_objects, compile)
191
192 # Compile model.
--> 193 model.compile(**saving_utils.compile_args_from_training_config(
194 training_config, custom_objects))
195 saving_utils.try_build_compiled_arguments(model)

E:\TF_env\lib\site-packages\tensorflow\python\keras\saving\saving_utils.py in compile_args_from_training_config(training_config, custom_objects)
209 with generic_utils.CustomObjectScope(custom_objects):
210 optimizer_config = training_config['optimizer_config']
--> 211 optimizer = optimizers.deserialize(optimizer_config)
212
213 # Recover losses.

E:\TF_env\lib\site-packages\tensorflow\python\keras\optimizers.py in deserialize(config, custom_objects)
863 if config['class_name'].lower() in all_classes:
864 config['class_name'] = config['class_name'].lower()
--> 865 return deserialize_keras_object(
866 config,
867 module_objects=all_classes,

E:\TF_env\lib\site-packages\tensorflow\python\keras\utils\generic_utils.py in deserialize_keras_object(identifier, module_objects, custom_objects, printable_module_name)
352
353 if 'custom_objects' in arg_spec.args:
--> 354 return cls.from_config(
355 cls_config,
356 custom_objects=dict(

E:\TF_env\lib\site-packages\tensorflow\python\keras\mixed_precision\experimental\loss_scale_optimizer.py in from_config(cls, config, custom_objects)
432 config = config.copy() # Make a copy, since we mutate config
433 config['optimizer'] = optimizers.deserialize(
--> 434 config['optimizer'], custom_objects=custom_objects)
435 config['loss_scale'] = keras_loss_scale_module.deserialize(
436 config['loss_scale'], custom_objects=custom_objects)

KeyError: 'optimizer'
`

Don't know anything about this error. the model is the same I download it from that notebook above.
please help me with how I run the model on the local machine.

[mrdbourke]

Hey Osprey,

I think your error (KeyError: 'optimizer') is coming from this line:

# saving the model again
model.save('effi_080_second.h5', include_optimizer=False)

Any reason not to include the optimizer?

When your model tries to load in, there's no optimizer so it errors.

Saving the model with the optimizer may fix your issue.

---

Some other things.

Did you have any code before:

model = tf.keras.models.load_model('model_0_80.h5')

When trying to load it on your local machine?

If you're trying to load EfficientNet on a local machine, you may have to try loading it with Custom Objects, for example:

new_model = tf.keras.models.load_model('model.h5', custom_objects={'CustomLayer': CustomLayer})

See this answer on Stack Overflow for more: https://stackoverflow.com/a/62326857/7900723

Does it work with the SavedModel format?

[Osprey-DS]

yes, Daniel, i did that as you have shown, I define include_optimizer = False and then I try It works.
Thank you Daniel for your help.
Once i will deploy the web app for this project, i will let you know on discord.
Thanks again for your time and help.