Saving pipelines (joblib, pickle) in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

What is the output of this pipeline saving code?

Consider the following Python code that trains a simple pipeline and saves it using joblib. What will be the output when loading and predicting with the saved pipeline?

ML Python

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression
import joblib

pipeline = Pipeline([
    ('scaler', StandardScaler()),
    ('clf', LogisticRegression())
])

X_train = [[0, 0], [1, 1], [2, 2], [3, 3]]
y_train = [0, 0, 1, 1]

pipeline.fit(X_train, y_train)
joblib.dump(pipeline, 'model.joblib')

loaded_pipeline = joblib.load('model.joblib')
pred = loaded_pipeline.predict([[1.5, 1.5]])
print(pred[0])

A 1

B 0

C Raises FileNotFoundError

D Raises AttributeError

💡 Hint

Think about what the model predicts for input close to training samples labeled 1.

The pipeline is trained on points where [2, 2] and [3, 3] are labeled 1. The input [1.5, 1.5] is closer to these points after scaling, so the logistic regression predicts class 1. The pipeline is saved and loaded correctly with joblib, so no errors occur.

❓ Model Choice

intermediate

Which method is best for saving a scikit-learn pipeline?

You have a trained scikit-learn pipeline. Which method is recommended to save and later reload the entire pipeline with minimal hassle?

A Use <code>pickle.dump()</code> and <code>pickle.load()</code> to save and load the pipeline.

B Save the pipeline as a CSV file.

C Save only the pipeline parameters as JSON and reconstruct the pipeline manually.

D Use <code>joblib.dump()</code> and <code>joblib.load()</code> to save and load the pipeline.

💡 Hint

Consider which method is optimized for large numpy arrays inside models.

Joblib is optimized for saving scikit-learn models and pipelines because it efficiently handles large numpy arrays. Pickle works but is slower and less efficient. Saving as JSON or CSV is not suitable for complex objects like pipelines.

❓ Hyperparameter

advanced

Which hyperparameter affects pipeline saving compatibility?

When saving a scikit-learn pipeline with joblib, which hyperparameter setting in the pipeline's components can cause issues when loading the saved pipeline in a different environment?

A Setting <code>random_state</code> to a fixed integer.

B Using custom transformer classes not defined in the loading environment.

C Setting <code>verbose=True</code> in pipeline steps.

D Using <code>n_jobs</code> with a value other than 1.

💡 Hint

Think about what happens if the code that defines a custom class is missing when loading.

Custom transformer classes must be available in the environment when loading a saved pipeline. If they are missing, loading raises errors. Other hyperparameters like random_state or n_jobs do not affect loading compatibility.

🔧 Debug

advanced

Why does this pipeline loading code raise an error?

Given the code below, why does loading the saved pipeline raise an error?

ML Python

import joblib

loaded_pipeline = joblib.load('saved_pipeline.pkl')
pred = loaded_pipeline.predict([[0, 0]])
print(pred)

A The file 'saved_pipeline.pkl' does not exist in the current directory.

B The file was saved with pickle, not joblib, causing incompatibility.

C The pipeline was saved with joblib but the file extension is incorrect.

D The pipeline was saved with a different Python version causing incompatibility.

💡 Hint

Check if the file path and name are correct and the file exists.

If the file 'saved_pipeline.pkl' does not exist, joblib.load raises FileNotFoundError. The file extension does not affect loading if the file exists. Pickle and joblib files are compatible if saved and loaded correctly. Python version differences usually cause warnings, not immediate errors.

🧠 Conceptual

expert

What is a key advantage of using joblib over pickle for saving ML pipelines?

Why is joblib often preferred over pickle for saving machine learning pipelines that include large numpy arrays?

A Joblib converts pipelines to JSON format for better readability.

B Joblib encrypts the saved files for security, unlike pickle.

C Joblib compresses data automatically and supports memory mapping for faster loading of large arrays.

D Joblib saves pipelines as plain text files for easier debugging.

💡 Hint

Think about performance and file size when saving large data.

Joblib uses efficient compression and supports memory mapping, which speeds up loading large numpy arrays inside pipelines. Pickle does not provide these features. Joblib does not encrypt or convert to JSON or plain text.