Certified Machine Learning ProfessionalDemo

A Machine Learning Engineer is configuring a cluster for a deep learning training run, with several settings to select. The cluster will be reused by other engineers on the team, and their datasets range in size from hundreds of MBs to hundreds of GBs. They must select a configuration that supports performant, stable deep learning training without excessive cost.

Which configuration accomplishes this?

A machine learning engineer is migrating a machine learning pipeline to Databricks Machine Learning. The pipeline must automatically refresh its model whenever it runs. The project is attached to the existing model_name model in the MLflow Model Registry.

They use the following code block as part of their solution:

Which statement explains the effect of the registered_model_name=model_name parameter and argument, given that model_name already exists in the MLflow Model Registry?

Which of the following lists all model stages available in the MLflow Model Registry?

A Data Scientist is creating a pipeline to train multiple machine-learning models. They have a dataframe containing a group column, many feature columns, and one label column.

Their requirements are:

• An independent model for each group
• Direct use of the data in the columns without transformation
• Flexibility to select from many different model types
• Efficient distributed-computing capabilities

They are deciding between the pyspark.ml library and pandas UDFs through groupby().ApplyInPandas() to train the models. They choose the groupBy().ApplyInPandas() approach.

What is a valid reason for this choice?

A data scientist has developed a model and calculated the model's RMSE on the test set. They assigned this value to the rmse variable. They now want to manually log the RMSE value in the MLflow run.

They write the following incomplete code block:

Which of the following lines of code can fill the blank so that the code block successfully completes the task?

A Machine Learning Engineer has defined a model with Python notebooks in a Databricks Git folder and wants to add unit tests to their CI/CD workflow. They need a workflow that is highly automated and straightforward to maintain.

Which strategy should they use?

A Data Scientist created a sales-forecasting sales-prediction model. The model is deployed to a model-serving endpoint with a schema. They need to call this model from Python and would prefer an SDK function to submit the request.

Which method meets these requirements?

A team is developing a machine learning pipeline that includes feature engineering, model training, model evaluation, and model deployment. During the last year, the project has grown substantially, with multiple contributors working on it. They currently test their pipelines and functions manually, but they want to establish an MLOps process and organize their unit tests.

Which approach should they use to organize functions and unit tests?

A Machine Learning Engineer is working with a Spark DataFrame that contains 100 million retail-transaction rows across thousands of stores. Each store needs its own demand-forecasting model, using the same scikit-learn pipeline. The engineer wants to train and apply these models in parallel for every store without collecting the data to the driver.

Which approach accomplishes this?

A Machine Learning Engineer requires a continuous deployment pipeline for models hosted on Databricks Model Serving. The deployment automation must run after a model is trained and registered with MLflow. Its objective is to deploy the latest model version from the MLflow Model Registry only when the model can satisfy the company’s strict latency requirement (P95 < 300ms) while handling production traffic.

How can the engineer confirm that new models satisfy the latency requirement when they are served in production?

A machine learning engineer wants to log and deploy a model as an MLflow pyfunc model. They have custom preprocessing that must be performed on feature variables before fitting the model or generating predictions with it. They choose to wrap this preprocessing in a custom model class, ModelWithPreprocess, where preprocessing occurs when fit and predict are called. They then log the fitted ModelWithPreprocess model as a pyfunc model.

Which of the following is a benefit of this approach when the logged pyfunc model is loaded for downstream deployment?

Which of the following statements defines label drift?

A Data Scientist is training machine-learning models in a notebook and must track varying parameters, a key performance metric named team_metric, and a summary graphic using MLflow. They find that tracking is not working as expected and review their logging code. For the logging portion of their code, they have:

Why is the code not logging according to their requirements?

A machine learning engineer has implemented numeric-drift monitoring by examining trends in the summary statistics of input variables. However, the engineer’s stakeholders want a more robust monitoring solution.

Which of the following can provide a more robust drift-monitoring solution for numeric feature variables?

A Data Scientist must build a sales-forecasting model that predicts each store’s sales monthly. They have been asked to configure Lakehouse Monitoring so the model can be assessed daily for potential accuracy metrics, such as RMSE. To do this, they will first configure Lakehouse Monitoring with the Inference profile type.

What is the next step in this process?

A machine learning engineer needs to view all active MLflow Model Registry Webhooks for a particular model. They are using this code block:

Which change must the machine learning engineer make to this code block for it to successfully complete the task?

A machine learning engineer is migrating a machine learning pipeline to use Databricks Machine Learning. They have programmatically identified the best run from an MLflow Experiment and saved its URI in the model_uri variable and its Run ID in the run_id variable. They have also established that the model was logged under the name "model".

The machine learning engineer now wants to register that model in the MLflow Model Registry with the name "best_model".

Which of the following code lines can they use to register the model in the MLflow Model Registry?

Which of the following Feature Store Client fs operations can return a Spark DataFrame for a data set associated with a Feature Store table?

A Machine Learning Engineer is training a large-scale gradient-boosting model with SparkML on a cluster of machines. After several iterations, the training job fails because memory overflows on one executor node. Cluster resources are limited to executor nodes that each have 16 CPU cores and 64 GB of RAM. The engineer wants to continue model training without changing hyperparameters or reducing the dataset size. They understand Spark’s architecture well and want to leverage its advantages.

Which approach will enable the Machine Learning Engineer to resolve this issue?

A Machine Learning Engineer is developing a script to score a large batch of customer data with a SparkML RandomForestClassificationModel registered in Unity Catalog. They want to follow commonly accepted coding best practices for this model.

Which approach best meets their needs?

A data scientist needs to track runs for a random forest model. Across each run, the scientist changes the number of trees and the maximum tree depth in the forest.

They write the following code:

Which Python object type must params be an instance of?

A Data Scientist is building a fraud-detection model with a massive, imbalanced dataset of more than 100 terabytes. They selected SparkML’s Logistic Regression and must optimize its hyperparameters. Because of the dataset’s size and the need for robust evaluation, they are evaluating strategies for hyperparameter tuning and model selection within the Spark ecosystem. They require the most computationally and time-efficient method to tune the SparkML Logistic Regression model and assess its performance.

Which approach accomplishes this?

A machine learning engineer needs to load data from the first version of a Delta table at a location path.

Which of the following code lines can be used to complete this task?

Which tool can support real-time deployments by packaging software together with its own application, tools, and libraries?