Saving and loading models in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

What is the output of this model saving code?

Consider the following code that trains a simple model and saves it. What will be printed after saving?

ML Python

import tensorflow as tf
from tensorflow.keras import layers

model = tf.keras.Sequential([
    layers.Dense(10, activation='relu', input_shape=(5,)),
    layers.Dense(1)
])
model.compile(optimizer='adam', loss='mse')

import numpy as np
x = np.random.random((10, 5))
y = np.random.random((10, 1))
model.fit(x, y, epochs=1, verbose=0)

model.save('my_model')
print('Model saved successfully')

A

Model saved successfully
Warning: Overwriting existing model

B No output

C Error: model.save() requires a file extension

D Model saved successfully

💡 Hint

The model.save() method saves the model in TensorFlow SavedModel format by default if no extension is given.

The code trains a simple model and saves it to the folder 'my_model'. Since no error occurs, the print statement outputs 'Model saved successfully'.

❓ Model Choice

intermediate

Which method correctly loads a saved Keras model?

You have saved a Keras model using model.save('my_model'). Which code snippet correctly loads this model?

A loaded_model = tf.keras.models.load_model('my_model')

B loaded_model = tf.keras.load('my_model')

C loaded_model = tf.load_model('my_model')

D loaded_model = tf.keras.models.Model.load('my_model')

💡 Hint

Look for the official Keras function to load models.

The correct function to load a saved Keras model is tf.keras.models.load_model(path). Other options are invalid or do not exist.

❓ Hyperparameter

advanced

Which saving format preserves custom objects in Keras models?

You have a Keras model with a custom layer. You want to save and load it without losing the custom layer. Which saving format should you use?

A Save model architecture as JSON only

B Save as HDF5 file with .h5 extension

C Save weights only with model.save_weights()

D Save as TensorFlow SavedModel format (folder)

💡 Hint

Custom objects are best preserved in one specific format.

Saving as HDF5 (.h5) preserves custom layers and objects better than SavedModel format, which may require extra steps to load custom objects.

❓ Metrics

advanced

After loading a saved model, which metric will NOT be available automatically?

You save a compiled Keras model with accuracy metric and load it later. Which metric will NOT be available without recompiling?

A Custom metric defined outside the model

B Loss function

C Model weights

D Accuracy metric

💡 Hint

Think about metrics that require custom code to be recognized.

Custom metrics defined outside the model are not saved automatically and require passing custom_objects when loading or recompiling.

🔧 Debug

expert

Why does this code raise an error when loading a model?

You saved a model with a custom activation function but get an error when loading it with tf.keras.models.load_model('my_model'). Why?

ML Python

def custom_activation(x):
    return tf.nn.relu(x) * tf.math.sigmoid(x)

model = tf.keras.Sequential([
    tf.keras.layers.Dense(10, activation=custom_activation, input_shape=(5,)),
    tf.keras.layers.Dense(1)
])
model.save('my_model')

# Later loading
loaded_model = tf.keras.models.load_model('my_model')

A The model was not compiled before saving

B The input_shape is missing in the loading code

C The custom activation function is not passed in custom_objects during loading

D The model.save() method does not support custom layers

💡 Hint

Custom functions must be provided when loading models that use them.

When loading a model with custom functions, you must pass them in the custom_objects argument to load_model, otherwise it raises an error.