Free Databricks-Machine-Learning-Associate: Databricks Certified Machine Learning Associate Exam Questions and Answers

A data scientist logged a model to an MLflow run under the artifact path "model". They want to register this model in the Unity Catalog registry as price_model in the retail schema of the ml_prod catalog. Which code block accomplishes this?

• AA. mlflow.set_registry_uri("databricks-uc") mlflow.register_model("runs:/<run_id>/model", "retail.price_model")
• BB. mlflow.set_tracking_uri("databricks-uc") mlflow.register_model("runs:/<run_id>/model", "price_model")
• CC. mlflow.set_registry_uri("databricks-uc") mlflow.register_model("runs:/<run_id>/model", "ml_prod.retail.price_model")
• DD. mlflow.set_registry_uri("databricks") mlflow.register_model("runs:/<run_id>/model", "ml_prod.retail.price_model")

A data scientist has a FeatureEngineeringClient named fe and a Spark DataFrame features_df. They want to create a feature table named customer_stats in the features schema of the main catalog, with customer_id as the primary key, and populate it with the rows of features_df. Which call does this?

• AA. fe.create_table(name="customer_stats", primary_keys=["customer_id"], df=features_df, description="Customer features")
• BB. fe.create_table(name="main.features.customer_stats", primary_keys=["customer_id"], df=features_df)
• CC. fe.write_table(name="main.features.customer_stats", df=features_df, mode="overwrite")
• DD. fe.create_table(name="main.features.customer_stats", df=features_df, description="Customer features")

A data scientist needs to predict whether a customer will churn using 40 numeric and categorical features. The relationships between the features and the label are nonlinear and involve interactions between features. The data scientist does not want to scale the features and needs per-feature importance scores. Which algorithm is the most appropriate?

• AA. k-nearest neighbors classifier with Euclidean distance
• BB. Linear support vector machine on standardized features
• CC. Gradient-boosted tree classifier
• DD. Logistic regression with L2 regularization on the original features

A data scientist is exploring a continuous feature named income and a categorical feature named region that has five distinct values. They want to compare the median and spread of income across the regions. Which visualization should they create?

• AA. Side-by-side box plots of income, one for each region
• BB. A pie chart of average income with one slice for each region
• CC. A scatter plot of income against region encoded as integers
• DD. A single histogram of income computed over all rows

A data scientist writes an objective function that trains a model and returns the validation accuracy. They then run fmin(fn=objective, space=search_space, algo=tpe.suggest, max_evals=20). The hyperparameters returned by fmin produce the lowest validation accuracy of all trials. Which change corrects this behavior?

• AA. Replace algo=tpe.suggest with algo=rand.suggest
• BB. Increase max_evals from 20 to 100
• CC. Return the negative of the validation accuracy from the objective function
• DD. Wrap each hyperparameter in hp.choice so values are sampled from lists

A data scientist wants a Spark DataFrame that contains only the 25th, 50th and 75th percentile rows for every numeric column of spark_df. Which code block does this?

• AA. spark_df.describe().filter(F.col("summary").isin("25%", "50%", "75%"))
• BB. spark_df.summary("25%", "50%", "75%")
• CC. spark_df.summary("quartiles")
• DD. spark_df.describe("25%", "50%", "75%")

An MLflow experiment contains 30 runs. Each run logged the parameters learning_rate and max_depth and the metric val_rmse. A data scientist wants to visualize how combinations of the two parameter values relate to val_rmse across the runs. Which part of the MLflow UI should they use?

• AA. The Artifacts tab on the page of the run with the lowest val_rmse
• BB. The parallel coordinates plot in the compare view for the selected runs
• CC. The system metrics chart on the page of a single run
• DD. The lineage tab on a Unity Catalog model version page

A data scientist is inside an active MLflow run. They want to record the final validation RMSE as a single numeric value and also save a residual plot, previously written to the local file residuals.png, with the run. Which pair of MLflow functions should they call?

• AA. mlflow.log_param and mlflow.log_dict
• BB. mlflow.set_tag and mlflow.log_input
• CC. mlflow.log_metric and mlflow.log_model
• DD. mlflow.log_metric and mlflow.log_artifact

A data scientist has logged a scikit-learn model with MLflow and stored its location in the variable model_uri. They have a pandas DataFrame named new_df with 50,000 rows that easily fits in driver memory, and they want to generate predictions for it. Which two approaches will produce the predictions? (Choose two.)

Select 2 answers.

• AA. Load the model with mlflow.sklearn.load_model(model_uri) and call transform(new_df) on the loaded model
• BB. Wrap the model with mlflow.pyfunc.spark_udf and call new_df.withColumn on the pandas DataFrame
• CC. Load the model with mlflow.sklearn.load_model(model_uri) and call predict(new_df) on the loaded model
• DD. Load the model with mlflow.pyfunc.load_model(model_uri) and call predict(new_df) on the loaded model

A machine learning engineer has an MLflow experiment whose ID is stored in exp_id. Each run logs a metric named val_rmse and a parameter named max_depth. Several runs failed partway through and logged only partial metrics. The engineer wants the single run with the lowest val_rmse among only the runs that completed successfully, using an MlflowClient named client. Which call returns that run?

• AA. client.search_runs([exp_id], filter_string="attributes.status = 'FINISHED'", order_by=["metrics.val_rmse DESC"], max_results=1)
• BB. client.search_runs([exp_id], filter_string="attributes.status = 'FINISHED'", order_by=["params.max_depth ASC"], max_results=1)
• CC. client.search_runs([exp_id], filter_string="attributes.status = 'FINISHED'", order_by=["metrics.val_rmse ASC"], max_results=1)
• DD. client.search_runs([exp_id], order_by=["metrics.val_rmse ASC"], max_results=1)