Error in loading auto keras pre-trained model

[krishdb38]

Hello,
auto-Keras prediction is very good in my data sets.
I trained (tuned) my models for several days and save the best model using

save the best performing model to file

model.save(f'model_auto_keras{max_trials}.h5')

the file is saved without any error.

when I am trying to load the saved file. using
model = tf.keras.models.load_model(model_path)
I am unable to load the model saying Value Error unknown layer; Custom > MultiCategoryEncoding .........

I searched a lot but can't find the MultiCategoryEncoding layer.
can anyone please help/share links with me to solve this problem?

[daraghimo]

just follow the tutorial here

import tensorflow as tf
from tensorflow.keras.models import load_model
import autokeras as ak

to save:
model.save("model_autokeras.h5")

to load:
loaded_model = load_model("model_autokeras", custom_objects=ak.CUSTOM_OBJECTS)

to use after loading
predicted_y = loaded_model.predict(tf.expand_dims(x_test, -1))

[gkourogiorgas]

I want to do something similar but without the weights. I used this:

from keras.models import model_from_json
model_json = model.export_model().to_json()
with open("model.json", "w") as json_file:
    json_file.write(model_json)

json_file = open('model.json', 'r')
loaded_model_json = json_file.read()
json_file.close()
loaded_model = model_from_json(loaded_model_json)

loaded_model.compile(loss='binary_crossentropy', optimizer='rmsprop', metrics=['accuracy'])

loaded_model.fit(train_, train['d'],
      epochs=100, verbose=0,
      validation_split=0.2, shuffle=False,
      class_weight=cw(train))

And I got this error:
Node: 'model/multi_category_encoding/string_lookup_218/None_Lookup/LookupTableFindV2' Table not initialized. [[{{node model/multi_category_encoding/string_lookup_218/None_Lookup/LookupTableFindV2}}]] [Op:__inference_train_function_43668109]

I think it needs the multi category encoding mapping. But I would expect it to re-run it when .fit is called

Any help?