Databricks Machine Learning Associate Question 243
Single answerYou are using Databricks AutoML to train a classification model on a dataset with three categorical features (A, B, C) and one numeric feature (D). Feature A has 4 unique values, feature B has 6 unique values, and feature C has 3 unique values. AutoML is set to perform hyperparameter tuning with 5-fold cross-validation. Assuming no additional feature engineering or filtering, how many models will be trained in total during the AutoML process?
- A
18
- B
180
- C
360
- D
540
Show answer and explanation
Correct answer: C
Explanation
To determine the total number of models trained, consider the combinations of categorical feature values (4 x 6 x 3 = 72) and the numeric feature used during training. Databricks AutoML performs hyperparameter tuning, and each unique hyperparameter configuration is evaluated using 5-fold cross-validation. Assuming one hyperparameter configuration per feature combination, the total is 72 combinations × 5 folds = 360 models.
- A. Incorrect.
This option underestimates the number of models. It does not properly account for the number of hyperparameter tuning iterations and cross-validation folds.
- B. Incorrect.
This option assumes only a single model per fold per hyperparameter configuration, but it does not account for all hyperparameter tuning iterations.
- C. Correct.
This option correctly calculates the total number of models trained. Each unique hyperparameter configuration is trained using 5-fold cross-validation, and Databricks AutoML typically explores multiple hyperparameter configurations during its search process. In this scenario, accounting for the categorical feature combinations (4 x 6 x 3 = 72) and 5-fold cross-validation, 360 models would be trained.
- D. Incorrect.
This option overestimates the number of models. It likely assumes additional models or configurations that are not part of the AutoML process.