A team is training a model and wants to log a custom metric that is computed from a validation set after each epoch. They are using a custom training loop in PyTorch and have enabled mlflow.pytorch.autolog(). They notice that autologging is logging the loss, but not their custom metric. What is the most likely reason and the correct fix?

Autologging for PyTorch only works with the Lightning framework; switch to mlflow.pytorch.lightning.autolog() to log custom metrics.

Autologging is disabled by default for PyTorch; call mlflow.pytorch.autolog(disable=False) to enable it.

The custom metric must be logged with mlflow.log_metric() before the training loop starts, otherwise autologging ignores it.

Autologging only logs metrics that are computed by the framework's built-in loss functions; manually call mlflow.log_metric() for the custom metric inside the training loop.

In a custom MLflow model class, where should real-time feature engineering logic be implemented so that it runs during inference?

In the log_model function of the model class

In the __init__ method of the model class

In the load_model function of the model class

In the predict method of the model class

A team is building a custom MLflow model that includes a preprocessing step: it needs to impute missing values in a numeric column with the median value computed during training. The median is stored as a float. How should the team make this median available to the model's predict method?

Save the median to a JSON file and load it from the current working directory in predict.

Log the median as a metric with mlflow.log_metric('median', median) and retrieve it in predict.

Hard-code the median value directly in the predict method's source code.

Pass the median as a parameter to the custom model's constructor and store it as an instance attribute in __init__.

DatabricksCertified Machine Learning Professional

Domain 1Objective 3

Advanced MLflow Usage MACHINE-LEARNING-PROFESSIONAL Practice Questions (Page 3)

Part of the Model Development domain, which accounts for 44% of the MACHINE-LEARNING-PROFESSIONAL exam. Databricks does not publish an official question count, but from its 120-minute exam (~50–80 total, ~22–35 in this domain), expect 6–9 from this objective — we provide 13 practice questions to prepare you well beyond it. (estimate)

13questions here

3free pages

3concepts

44%of the exam

Questions 11–13

11expert · hard
A team is training a model and wants to log a custom metric that is computed from a validation set after each epoch. They are using a custom training loop in PyTorch and have enabled mlflow.pytorch.autolog(). They notice that autologging is logging the loss, but not their custom metric. What is the most likely reason and the correct fix?
Select an answer first
12foundation · easy
In a custom MLflow model class, where should real-time feature engineering logic be implemented so that it runs during inference?
Select an answer first
13application · medium
A team is building a custom MLflow model that includes a preprocessing step: it needs to impute missing values in a numeric column with the median value computed during training. The median is stored as a float. How should the team make this median available to the model's predict method?
Select an answer first

Finished these 3 questions?

Review the revealed explanations, or continue through the curriculum.

No more pagesBack to MACHINE-LEARNING-PROFESSIONAL

Free Basic Practice is a study aid with revealable answers — not a scored exam. Examers.io is independent and not affiliated with or endorsed by Databricks. “MACHINE-LEARNING-PROFESSIONAL” is a trademark of its owner, used for identification only.

Advanced MLflow Usage MACHINE-LEARNING-PROFESSIONAL Practice Questions (Page 3)

Questions 11–13

In a custom MLflow model class, where should real-time feature engineering logic be implemented so that it runs during inference?

A team is building a custom MLflow model that includes a preprocessing step: it needs to impute missing values in a numeric column with the median value computed during training. The median is stored as a float. How should the team make this median available to the model's predict method?

Keep studying — Model Development