CODEWITHSUNDEEP
pythontensorflowkeras
hanzgs
hanzgs

Reputation: 1616

How to save keras recommenderNet model?

After building the model from Embedding RecommenderNet model, how to save it, link to the doc is https://keras.io/examples/structured_data/collaborative_filtering_movielens/

class RecommenderNet(keras.Model):
    def __init__(self, num_users, num_movies, embedding_size, **kwargs):
        super(RecommenderNet, self).__init__(**kwargs)
        self.num_users = num_users
        self.num_movies = num_movies
        self.embedding_size = embedding_size
        self.user_embedding = layers.Embedding(
            num_users,
            embedding_size,
            embeddings_initializer="he_normal",
            embeddings_regularizer=keras.regularizers.l2(1e-6),
        )
        self.user_bias = layers.Embedding(num_users, 1)
        self.movie_embedding = layers.Embedding(
            num_movies,
            embedding_size,
            embeddings_initializer="he_normal",
            embeddings_regularizer=keras.regularizers.l2(1e-6),
        )
        self.movie_bias = layers.Embedding(num_movies, 1)

    def call(self, inputs):
        user_vector = self.user_embedding(inputs[:, 0])
        user_bias = self.user_bias(inputs[:, 0])
        movie_vector = self.movie_embedding(inputs[:, 1])
        movie_bias = self.movie_bias(inputs[:, 1])
        dot_user_movie = tf.tensordot(user_vector, movie_vector, 2)
        # Add all the components (including bias)
        x = dot_user_movie + user_bias + movie_bias
        # The sigmoid activation forces the rating to between 0 and 1
        return tf.nn.sigmoid(x)


model = RecommenderNet(num_users, num_movies, EMBEDDING_SIZE)
model.compile(
    loss=tf.keras.losses.BinaryCrossentropy(), optimizer=keras.optimizers.Adam(lr=0.001)
)
history = model.fit(
    x=x_train,
    y=y_train,
    batch_size=64,
    epochs=5,
    verbose=1,
    validation_data=(x_val, y_val),
)

tried these

model.save('model.h5py')
tf.keras.models.save_model(model, overwrite=True, include_optimizer=True, save_format='h5')

Both throws

NotImplementedError: Saving the model to HDF5 format requires the model to be a Functional model or 
a Sequential model.It does not work for subclassed models, because such models are defined via the body of 
a Python method, which isn't safely serializable. Consider saving to the Tensorflow SavedModel 
format (by setting save_format="tf") or using `save_weights`.

model type is main.RecommenderNet

Upvotes: 0

Views: 1324

Answers (3)

hanzgs
hanzgs

Reputation: 1616

Actually recreating the model with

keras.models.load_model('path_to_my_model')

didn't work for me First we have to save_weights from the built model

model.save_weights('model_weights', save_format='tf')

Then we have to initiate a new instance for the subclass Model then compile and train_on_batch with one record and load_weights of built model

loaded_model = RecommenderNet(num_users, num_movies, EMBEDDING_SIZE)
loaded_model.compile(loss=tf.keras.losses.BinaryCrossentropy(), optimizer=keras.optimizers.Adam(lr=0.001))
loaded_model.train_on_batch(x_train[:1], y_train[:1])
loaded_model.load_weights('model_weights')

This work perfectly in TensorFlow==2.2.0

Upvotes: 1

hanzgs
hanzgs

Reputation: 1616

I upgraded tensorflow 1.14.0 to 2.2.0, save_model worked

tf.keras.models.save_model(model,'./saved_model')

But with the load_model, after loading the model, while prediction

loaded_model = tf.keras.models.load_model('./saved_model')

I get value error

rates = loaded_model.predict(user_product_array).flatten()

    ValueError: Python inputs incompatible with input_signature:
      inputs: (
        Tensor("IteratorGetNext:0", shape=(None, 2), dtype=int32))
      input_signature: (
        TensorSpec(shape=(None, 2), dtype=tf.int64, name='input_1'))

May I know the issue here?

Upvotes: 0

Jonas
Jonas

Reputation: 1769

So as the error says, you can use save_format="tf" since your model is not Functional or Sequential:

model.save('model.py')
tf.keras.models.save_model(model, overwrite=True, include_optimizer=True, save_format='tf')

Also as seen in the documentation you can use:

keras_model_path = "/tmp/keras_save"
model.save(keras_model_path)

Upvotes: 1

Related Questions