Saving and loading a model in Keras is a straightforward process. The model can be saved in two formats: the full architecture of the model, including the model weights, or just the model weights. In this article, we will cover both methods with code examples.

Saving the full architecture of the model

To save the full architecture of the model, you can use the save() method of the model object. The method takes the filepath where the model should be saved and a few additional optional arguments.

from keras.models import Sequential
from keras.layers import Dense

# Create a simple model
model = Sequential()
model.add(Dense(32, input_shape=(784,), activation='relu'))
model.add(Dense(10, activation='softmax'))

# Save the model to a file
model.save('model.h5')

In this example, we are creating a simple model with two dense layers and saving it to a file named 'model.h5'. The save() method will save the model in the HDF5 format, which is a standard format for storing large datasets. The file extension '.h5' is the default when saving a model in Keras.

Loading the full architecture of the model

To load the full architecture of the model, you can use the keras.models.load_model() function. The function takes the filepath of the saved model and returns the model object.

from keras.models import load_model

# Load the model from a file
model = load_model('model.h5')

Saving just the model weights

To save just the model weights, you can use the save_weights() method of the model object. The method takes the filepath where the weights should be saved and a few additional optional arguments.

# Create a model
model = Sequential()
model.add(Dense(32, input_shape=(784,), activation='relu'))
model.add(Dense(10, activation='softmax'))

# Save the model weights
model.save_weights('weights.h5')

Loading just the model weights

To load just the model weights, you can use the load_weights() method of the model object. The method takes the filepath of the saved weights and a few additional optional arguments.

# Create a model
model = Sequential()
model.add(Dense(32, input_shape=(784,), activation='relu'))
model.add(Dense(10, activation='softmax'))

# Load the model weights
model.load_weights('weights.h5')

It is worth noting that when you are loading the weights into a model, the model architecture should be the same as the one that was used to train and save the weights, otherwise, you will get a ValueError.

In conclusion, saving and loading models and weights in Keras is very simple, you can use the save() method to save the entire model, save_weights() method to save the weights only and load_model(), load_weights() method to load the model and weights respectively. It's important to check the architecture of the model before loading the weights, otherwise, it will raise an exception.

Using custom objects when loading a model

When loading a model that was saved with custom objects, such as custom layers or loss functions, you need to pass the custom objects to the load_model() function using the custom_objects argument. The custom_objects argument should be a dictionary that maps the names of the custom objects to the corresponding classes or functions.

For example, let's say you have a custom layer named 'MyLayer' and you want to use it when loading a model. Here's an example of how you would use the custom_objects argument:

# Define custom layer
class MyLayer(Layer):
    ...

# Create a model
model = Sequential()
model.add(MyLayer(input_shape=(784,)))
model.add(Dense(10, activation='softmax'))

# Save the model
model.save('model_with_custom_layer.h5')

# Load the model
loaded_model = load_model('model_with_custom_layer.h5', custom_objects={'MyLayer': MyLayer})

Saving the model in other formats

Keras also supports saving models in other formats, such as the TensorFlow SavedModel format, and the JSON and YAML formats.

To save a model in the TensorFlow SavedModel format, you can use the tf.saved_model.save() function from TensorFlow. The function takes the model, the directory where the model should be saved, and a few additional optional arguments.

import tensorflow as tf

# Create a model
model = Sequential()
model.add(Dense(32, input_shape=(784,), activation='relu'))
model.add(Dense(10, activation='softmax'))

# Save the model in TensorFlow SavedModel format
tf.saved_model.save(model, 'saved_model')

To save a model in the JSON or YAML format, you can use the to_json() or to_yaml() method of the model object, respectively. The methods return the model architecture as a string in the corresponding format. You can then save the string to a file or send it to another application.

# Create a model
model = Sequential()
model.add(Dense(32, input_shape=(784,), activation='relu'))
model.add(Dense(10, activation='softmax'))

# Save the model architecture in JSON format
json_string = model.to_json()
with open('model.json', 'w') as f:
    f.write(json_string)

# Save the model architecture in YAML format
yaml_string = model.to_yaml()
with open('model.yaml', 'w') as f:
    f.write(yaml_string)

Reloading a model from json/yaml

To reload the model from json or yaml file you can use the keras.models.model_from_json() or keras.models.model_from_yaml() function respectively. Both functions take the string representation of the model architecture and return the model object.

from keras.models import model_from_json
## Popular questions 
1. How can I save the full architecture of a model in Keras?

You can use the `save()` method of the model object to save the full architecture of a model in Keras. The method takes the filepath where the model should be saved and a few additional optional arguments. For example:

model.save('model.h5')

2. How can I load the full architecture of a model in Keras?

You can use the `load_model()` function to load the full architecture of a model in Keras. The function takes the filepath of the saved model and returns the model object. For example:

loaded_model = load_model('model.h5')

3. How can I save just the model weights in Keras?

You can use the `save_weights()` method of the model object to save just the model weights in Keras. The method takes the filepath where the weights should be saved and a few additional optional arguments. For example:

model.save_weights('weights.h5')

4. How can I load just the model weights in Keras?

You can use the `load_weights()` method of the model object to load just the model weights in Keras. The method takes the filepath of the saved weights and a few additional optional arguments. For example:

model.load_weights('weights.h5')

5. How can I use custom objects when loading a model in Keras?

When loading a model that was saved with custom objects, such as custom layers or loss functions, you need to pass the custom objects to the `load_model()` function using the `custom_objects` argument. The `custom_objects` argument should be a dictionary that maps the names of the custom objects to the corresponding classes or functions. For example:

loaded_model = load_model('model_with_custom_layer.h5', custom_objects={'MyLayer': MyLayer})

### Tag 
Serialization