Training Pipelines and Cross-Validation MACHINE-LEARNING-ASSOCIATE Practice Questions (Page 2)

Questions 6–10

1. 6expert · hard

   A team is using GridSearchCV to tune a GradientBoostingClassifier. They have a parameter grid with 4 values for learning_rate and 5 values for max_depth. They set cv=5. After the search, they want to report the model's expected performance. What is the most appropriate score to report?

   • AThe best_score_ attribute from GridSearchCV.
   • BThe score of the best model on the first fold only.
   • CThe average of all 20 combination scores.
   • DThe score of the best model on the full training data.
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   What is the primary purpose of a validation set in a train-validation split?

   • ATo test the model on completely new data after deployment
   • BTo train the model on additional data
   • CTo increase the size of the training set
   • DTo evaluate the model's performance on unseen data during development
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A data scientist is building a churn prediction model. They need to standardize numeric features and one-hot encode categorical features before training a logistic regression. They also want to ensure the same preprocessing steps are applied to the validation set without data leakage. What is the most maintainable approach?

   • ASplit the data, then use MLflow to log the raw training and validation DataFrames without preprocessing.
   • BFit a StandardScaler and OneHotEncoder on the full dataset, then transform both training and validation sets separately.
   • CManually apply StandardScaler and OneHotEncoder to the training set, then independently fit new scalers on the validation set.
   • DUse scikit-learn's Pipeline with ColumnTransformer to chain preprocessing and the logistic regression, then call fit on the training data only.
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A data scientist is using MLflow to track a training pipeline. They have a preprocessing step that imputes missing values and a model training step. They want to log the model along with the preprocessing steps so that the entire pipeline can be loaded and used for inference later. What should they do?

   • ALog only the trained model with mlflow.sklearn.log_model.
   • BLog the preprocessing steps and the model as separate artifacts.
   • CCreate a scikit-learn Pipeline that includes preprocessing and the model, then log the entire pipeline with mlflow.sklearn.log_model.
   • DSave the raw data and the model, then re-run preprocessing during inference.
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A data scientist is building a training pipeline for a text classification task. The pipeline includes TF-IDF vectorization, a classifier, and evaluation. They want to ensure that the TF-IDF vocabulary is learned only from the training data to avoid leakage. How should they structure the pipeline?

   • AFit TF-IDF on the validation set and transform the training set.
   • BFit TF-IDF on the full dataset, then split into train and validation.
   • CUse a Pipeline with TF-IDF and the classifier, and fit the pipeline on the full dataset.
   • DSplit the data first, then fit TF-IDF on the training set and transform both training and validation sets.
   Reveal Solution Discussion Select an answer first