Use One-Hot Encoding for Categorical Features MACHINE-LEARNING-ASSOCIATE Practice Questions (Page 2)

Questions 6–10

1. 6expert · hard

   A machine learning engineer is deploying a Spark ML pipeline for real-time inference. The training data had a categorical column 'device_type' with values 'mobile', 'tablet', 'desktop'. During inference, a new device type 'smartwatch' appears. The pipeline currently uses StringIndexer with default settings. What will happen when the pipeline processes the new value?

   • AThe pipeline will assign the new value to a special 'unknown' category and continue processing.
   • BThe pipeline will map the new value to the most frequent category from training.
   • CThe pipeline will treat the new value as a missing value and impute it with the mode.
   • DThe pipeline will throw an error because StringIndexer cannot handle unseen labels by default.
   Reveal Solution Discussion Select an answer first
2. 7application · easy

   A data analyst is explaining one-hot encoding to a colleague. They have a categorical feature 'day_of_week' with values 'Mon', 'Tue', 'Wed', 'Thu', 'Fri'. Which statement accurately describes one-hot encoding for this feature?

   • AIt creates a single column with the frequency of each day in the dataset.
   • BIt creates four binary columns, with the fifth day represented by all zeros.
   • CIt creates five binary columns, each indicating the presence of a specific day, with exactly one column equal to 1 for each row.
   • DIt creates a single column with values 1 through 5 representing the day of the week.
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A machine learning engineer is preparing a dataset with a 'color' column containing values 'red', 'green', 'blue'. They plan to use one-hot encoding. During model inference, they encounter a new value 'yellow' that was not present in the training data. What is the most appropriate way to handle this unseen category?

   • AIgnore the unseen category and let the model treat it as a missing value, which is acceptable for tree-based models.
   • BSet the StringIndexer handleInvalid parameter to 'keep' and use a OneHotEncoder that can represent unseen categories as a separate indicator.
   • CRetrain the model every time a new category appears, ensuring all possible categories are always included.
   • DMap all unseen categories to the most frequent category from training, such as 'red'.
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A data scientist is comparing one-hot encoding with label encoding for a categorical feature 'city' with 50 categories in a linear regression model. What is the primary advantage of one-hot encoding over label encoding in this context?

   • AOne-hot encoding preserves the ordinal relationship among categories, which label encoding does not.
   • BOne-hot encoding is always more interpretable than label encoding for tree-based models.
   • COne-hot encoding reduces the dimensionality of the feature space compared to label encoding.
   • DOne-hot encoding avoids imposing a false ordinal relationship, which could mislead a linear model.
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A data scientist is building a Spark ML pipeline to predict employee attrition. The dataset has a categorical column 'job_role' with 15 unique values. They want to one-hot encode this column. What is the correct way to prepare this column for one-hot encoding?

   • AApply OneHotEncoder directly to the job role strings.
   • BUse a custom UDF to manually create binary columns for each job role.
   • CUse VectorAssembler to combine the job role strings into a single vector, then apply OneHotEncoder.
   • DUse StringIndexer to convert the job role strings to numeric indices, then apply OneHotEncoder.
   Reveal Solution Discussion Select an answer first