Model versioning in TensorFlow - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why use model versioning in machine learning?

Imagine you train a model and later improve it. Why is it important to keep versions of your models?

A To avoid using any validation data

B To make the model training faster

C To reduce the size of the model files

D To track changes and compare performance between different models

💡 Hint

Think about how you keep different drafts of a document to see which one is better.

Model versioning helps keep track of changes and allows comparison between different model versions to choose the best one.

❓ Predict Output

intermediate

What is the output of this TensorFlow model save code?

Consider this code saving a model with versioning:

import tensorflow as tf
model = tf.keras.Sequential([tf.keras.layers.Dense(1, input_shape=(3,))])
model.save('models/v1')

What will be created in the filesystem?

TensorFlow

import tensorflow as tf
model = tf.keras.Sequential([tf.keras.layers.Dense(1, input_shape=(3,))])
model.save('models/v1')

A An error because the path is a folder, not a file

B A single file named 'models/v1.h5' saved in the current directory

C A folder named 'models/v1' containing the saved model files

D Nothing saved because the model is not compiled

💡 Hint

TensorFlow saves models in folders when given a directory path.

When saving with a folder path, TensorFlow creates a folder with the model files inside. The model does not need to be compiled to save.

❓ Hyperparameter

advanced

Choosing the right version number for a new model

You have saved model versions 'v1', 'v2', and 'v3'. You retrain the model with more data and want to save the new model. Which version number should you use?

A v2, to overwrite the previous model with the same data

B v4, because it is the next sequential version

C v1, to keep the version numbers low

D v10, to skip numbers and confuse users

💡 Hint

Think about how software versions increase logically.

Using the next sequential version number (v4) keeps versioning clear and organized.

❓ Metrics

advanced

Comparing model versions using accuracy

You have two model versions saved: v1 with accuracy 0.85 and v2 with accuracy 0.83. Which model should you deploy?

A v1, because it has higher accuracy

B v2, because lower accuracy means faster predictions

C Neither, because accuracy is not important

D v2, because it is the latest version regardless of accuracy

💡 Hint

Higher accuracy usually means better predictions.

Deploying the model with higher accuracy (v1) is better for prediction quality.

🔧 Debug

expert

Why does loading a saved model version fail?

You saved a TensorFlow model with model.save('models/v1'). Later, you try to load it with tf.keras.models.load_model('models/v2') and get an error. Why?

A Because 'models/v2' folder does not exist or has no saved model

B Because the model was saved with the wrong TensorFlow version

C Because the model was not compiled before saving

D Because the load_model function requires a file, not a folder

💡 Hint

Check if the folder you are loading from actually exists.

Loading fails because 'models/v2' does not exist or does not contain a saved model. The path must match the saved model folder.