Sentiment analysis with scikit-learn in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Model Choice

intermediate

Choosing the best model for sentiment analysis

You want to build a sentiment analysis model using scikit-learn on a dataset of movie reviews labeled positive or negative. Which model is most suitable for this binary text classification task?

A LinearSVC (Support Vector Classifier with linear kernel)

B KMeans clustering

C LinearRegression

D DBSCAN clustering

💡 Hint

Think about models designed for classification, not clustering or regression.

LinearSVC is a linear classifier suitable for binary classification tasks like sentiment analysis. KMeans and DBSCAN are clustering algorithms, not classifiers. LinearRegression is for continuous output, not categories.

❓ Predict Output

intermediate

Output of text vectorization step

What is the shape of the feature matrix X after applying CountVectorizer to 1000 text reviews, if the vocabulary size is 5000?

ML Python

from sklearn.feature_extraction.text import CountVectorizer
texts = ['sample text data'] * 1000
vectorizer = CountVectorizer(max_features=5000)
X = vectorizer.fit_transform(texts)
print(X.shape)

A (5000, 1000)

B (1000, 5000)

C (1000, 1)

D (1, 5000)

💡 Hint

Rows represent samples, columns represent features.

CountVectorizer transforms text into a matrix where each row is a sample and each column is a word feature. With 1000 samples and 5000 features, shape is (1000, 5000).

❓ Hyperparameter

advanced

Choosing the right hyperparameter for LogisticRegression

You train a LogisticRegression model for sentiment analysis. Which hyperparameter controls the strength of regularization to prevent overfitting?

A penalty (Type of regularization)

B max_iter (Maximum iterations for solver)

C C (Inverse of regularization strength)

D solver (Algorithm to use in optimization)

💡 Hint

Look for the parameter that adjusts how much the model avoids complexity.

The 'C' parameter in LogisticRegression is the inverse of regularization strength. Smaller values specify stronger regularization, helping to reduce overfitting.

❓ Metrics

advanced

Interpreting classification report metrics

After training a sentiment classifier, you get these metrics for the positive class: precision=0.8, recall=0.5, f1-score=0.62. What does the low recall indicate?

A The model is overfitting the training data

B The model predicts many false positives

C The model has perfect accuracy

D The model misses many positive reviews (false negatives are high)

💡 Hint

Recall measures how many actual positives are found.

Recall is the fraction of true positives found out of all actual positives. Low recall means many positive samples are missed (false negatives).

🔧 Debug

expert

Debugging unexpected accuracy drop after vectorizer change

You trained a sentiment model with CountVectorizer and got 85% accuracy. After switching to TfidfVectorizer with default settings, accuracy dropped to 70%. What is the most likely cause?

A TfidfVectorizer normalizes word counts, which may reduce the impact of frequent sentiment words

B TfidfVectorizer always produces fewer features than CountVectorizer, causing underfitting

C CountVectorizer applies stemming by default, but TfidfVectorizer does not

D TfidfVectorizer requires labels to be numeric, unlike CountVectorizer

💡 Hint

Think about how TF-IDF changes word importance compared to raw counts.

TfidfVectorizer scales down frequent words, which can reduce the weight of common sentiment words that were important in CountVectorizer, causing accuracy drop.