Feature union in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What does FeatureUnion do in a machine learning pipeline?

Imagine you want to combine different sets of features extracted from the same data before training a model. What is the main purpose of using FeatureUnion in this context?

A It combines multiple feature extraction processes by concatenating their outputs into a single feature set.

B It selects the best single feature extractor among many to use for the model.

C It reduces the dimensionality of features by applying PCA on the combined features.

D It splits the dataset into multiple parts to train separate models independently.

💡 Hint

Think about how you can use different ways to get features and then join them together before training.

FeatureUnion takes multiple feature extraction steps and concatenates their outputs side-by-side, creating a combined feature set for the model to learn from.

❓ Predict Output

intermediate

Output shape after FeatureUnion transformation

Given the following code, what is the shape of X_transformed?

ML Python

from sklearn.pipeline import FeatureUnion
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA
import numpy as np

X = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]])

fu = FeatureUnion([
    ('scale', StandardScaler()),
    ('pca', PCA(n_components=1))
])

X_transformed = fu.fit_transform(X)

A (3, 4)

B (3, 2)

C (3, 3)

D (3, 1)

💡 Hint

Consider the output dimensions of each transformer and how FeatureUnion combines them.

The input X has shape (3, 3). StandardScaler outputs shape (3, 3), while PCA(n_components=1) outputs shape (3, 1). FeatureUnion concatenates these outputs along the feature axis, resulting in shape (3, 4).

❓ Model Choice

advanced

Choosing transformers for FeatureUnion

You want to build a FeatureUnion that combines text features and numeric features for a classification task. Which combination of transformers is most appropriate?

A Use <code>PCA</code> for text features and <code>OneHotEncoder</code> for numeric features.

B Use <code>CountVectorizer</code> for numeric features and <code>MinMaxScaler</code> for text features.

C Use <code>TfidfVectorizer</code> for text and <code>StandardScaler</code> for numeric features.

D Use <code>LabelEncoder</code> for text and <code>Normalizer</code> for numeric features.

💡 Hint

Think about which transformers are designed for text and which for numeric data.

TfidfVectorizer converts text into numeric features based on word importance. StandardScaler normalizes numeric features. This combination is suitable for combining text and numeric data.

❓ Hyperparameter

advanced

Effect of n_jobs parameter in FeatureUnion

What is the effect of setting n_jobs=-1 in a FeatureUnion?

A It disables parallelism and runs transformers sequentially.

B It runs all transformers in parallel using all available CPU cores.

C It limits the number of features to one per transformer.

D It automatically tunes hyperparameters of each transformer.

💡 Hint

Consider what n_jobs=-1 usually means in scikit-learn.

Setting n_jobs=-1 tells scikit-learn to use all CPU cores to run transformers in parallel, speeding up the feature extraction process.

🔧 Debug

expert

Why does this FeatureUnion pipeline raise a ValueError?

Consider this code snippet:

from sklearn.pipeline import FeatureUnion
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA
import numpy as np

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

fu = FeatureUnion([
    ('scale', StandardScaler()),
    ('pca', PCA(n_components=3))
])

X_transformed = fu.fit_transform(X)

Why does this code raise a ValueError?

A The input array X must be 1-dimensional for FeatureUnion.

B StandardScaler cannot process arrays with less than 3 rows.

C FeatureUnion requires all transformers to output the same number of features.

D PCA is asked to extract more components (3) than the number of features (2) in X.

💡 Hint

Check the shape of X and the n_components parameter of PCA.

PCA cannot extract more components than the number of features in the input data. Here, X has 2 features, but PCA is set to extract 3 components, causing a ValueError.