Pipeline best practices in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why use a pipeline in machine learning?

Which of the following is the main reason to use a pipeline when building a machine learning model?

A To reduce the number of features in the dataset by default

B To increase the size of the training dataset automatically

C To combine data preprocessing and model training steps into one workflow

D To make the model run faster by skipping data cleaning

💡 Hint

Think about how pipelines help organize multiple steps in a machine learning task.

Pipelines help combine preprocessing and model training into a single, repeatable workflow. This ensures consistent data handling and easier experimentation.

❓ Predict Output

intermediate

Output of pipeline with scaling and logistic regression

What will be the output of the following code snippet?

ML Python

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression
import numpy as np

X = np.array([[1, 2], [3, 4], [5, 6]])
y = np.array([0, 1, 0])

pipe = Pipeline([
    ('scaler', StandardScaler()),
    ('clf', LogisticRegression(random_state=42))
])

pipe.fit(X, y)
pred = pipe.predict(np.array([[2, 3]]))
print(pred[0])

A 1

B 0

C IndexError

D ValueError

💡 Hint

Consider how the logistic regression model predicts based on the scaled input.

The pipeline scales the input [[2, 3]] and then predicts using the trained logistic regression model. Given the training data, the prediction is class 0.

❓ Model Choice

advanced

Choosing the right pipeline step for text data

You want to build a pipeline to classify text messages as spam or not spam. Which step should you add before the classifier to convert text into numbers?

A CountVectorizer()

B PCA()

C StandardScaler()

D KMeans()

💡 Hint

Think about how to convert text data into a format a model can understand.

CountVectorizer converts text into a matrix of token counts, which is suitable for feeding into classifiers. StandardScaler and PCA are for numeric data, and KMeans is a clustering algorithm.

❓ Hyperparameter

advanced

Setting hyperparameters in a pipeline

Given a pipeline with a scaler and a random forest classifier named 'clf', how do you set the number of trees (n_estimators) to 100 in the classifier using the pipeline object?

A pipe.set_params(clf__n_estimators=100)

B pipe.set_params(n_estimators=100)

C pipe.clf.n_estimators = 100

D pipe.set_params(scaler__n_estimators=100)

💡 Hint

Remember how to access parameters of steps inside a pipeline.

To set parameters of a step inside a pipeline, use the step name followed by two underscores and the parameter name. Here, 'clf__n_estimators' sets the number of trees in the classifier.

🔧 Debug

expert

Why does this pipeline cause a data leakage problem?

Consider this pipeline code:

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

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

pipe.fit(X_train, y_train)

# Later
X_train_scaled = pipe.named_steps['scaler'].transform(X_train)
X_test_scaled = pipe.named_steps['scaler'].transform(X_test)

model = LogisticRegression()
model.fit(X_train_scaled, y_train)

predictions = model.predict(X_test_scaled)

What is the main issue with this approach?

A The scaler is fit twice, causing data leakage from test data

B The logistic regression model is trained twice on the same data

C The scaler is fit only on training data, so no leakage occurs

D The pipeline is not used for prediction, causing inconsistent preprocessing

💡 Hint

Think about how the pipeline should be used for both training and prediction.

The pipeline is fit on training data, but then the scaler is used separately to transform data and a new logistic regression model is trained outside the pipeline. This breaks the pipeline's consistency and can cause errors or leakage if not careful. The pipeline should be used end-to-end for both training and prediction.