Machine Learning Mastery Quiz — Quiz

Linear models, ensembles, clustering, tuning, imbalance and deployment. · 10 questions

Question 1 of 10

Logistic regression converts a weighted sum into a probability using which function?

aReLUbSoftmax over featurescSigmoiddStep function

Question 2 of 10

Which scikit-learn model tends to set some coefficients exactly to zero?

aLassobRidgecLinearRegressiondKNeighborsRegressor

Question 3 of 10

A random forest reduces overfitting mainly by…

aTraining trees sequentially on residualsbAveraging many trees trained on bootstrap samples with random feature subsetscPruning a single treedScaling the features

Question 4 of 10

Why must features usually be scaled before k-means?

aK-means only accepts values between 0 and 1bScaling makes it run on GPUscScaling removes outliersdIt uses distances, so large-unit features would dominate

Question 5 of 10

What does this code keep?

pca = PCA(n_components=0.95)
Z = pca.fit_transform(X_scaled)

aExactly 95 componentsbEnough components to explain 95% of the variancecThe 95% of rows with least noisedComponents with variance above 0.95

Question 6 of 10

For a churn dataset with only 3% positives, which scoring is most informative during tuning?

aaccuracybr2caverage_precisiondneg_mean_squared_error

Question 7 of 10

Which splitter should you use when validating a weekly sales forecasting model?

aTimeSeriesSplitbKFold(shuffle=True)cStratifiedKFolddLeaveOneOut

Question 8 of 10

What is wrong with this workflow?

X_res, y_res = SMOTE().fit_resample(X, y)
X_tr, X_te, y_tr, y_te = train_test_split(X_res, y_res)

aSMOTE needs scaled databtrain_test_split cannot handle arrayscNothing, this is the standard approachdSynthetic samples derived from test rows leak into training, inflating scores

Question 9 of 10

In gradient boosting, lowering `learning_rate` from 0.3 to 0.03 usually requires…

aMore boosting iterations to reach similar performancebFewer iterationscRemoving early stoppingdDeeper trees only

Question 10 of 10

You `joblib.dump` a fitted pipeline with scikit-learn 1.5 and load it in production. What is the safest practice?

aLoad it with any newer scikit-learn versionbConvert it to CSV firstcPin the same library versions in production and only load files from trusted sourcesdRe-fit the pipeline on each request