Advanced MLflow Usage MACHINE-LEARNING-PROFESSIONAL Practice Questions (Page 1)

Questions 1–5

1. 1foundation · easy

   Which MLflow function is used to log a custom metric that changes over time, such as the loss at each training epoch?

   • Amlflow.log_artifact(local_path)
   • Bmlflow.log_param(key, value)
   • Cmlflow.log_text(text, artifact_file)
   • Dmlflow.log_metric(key, value, step=None)
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A data scientist is running an experiment and wants to log a list of evaluation metrics (precision, recall, F1) for multiple thresholds. They want to compare these across runs in the MLflow UI. Which approach is most appropriate?

   • ALog each metric-threshold combination as a separate metric with a name like 'precision@0.5' and the value, using mlflow.log_metric() for each.
   • BLog the metrics to a nested run for each threshold, with the threshold as the run name.
   • CLog the entire list of metrics as a single JSON artifact using mlflow.log_artifact().
   • DLog the metrics as parameters using mlflow.log_param() with the threshold in the name.
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A team is building a custom MLflow model that performs feature scaling inside its predict method. The scaler was fitted on training data and saved as a pickle file. The team wants to log the scaler as an artifact alongside the model so it can be loaded during inference. Which MLflow call correctly saves the scaler with the model?

   • Amlflow.log_param('scaler_path', 'scaler.pkl') and load the file from that path in predict.
   • Bmlflow.pyfunc.log_model(artifact_path='model', python_model=CustomModel(), artifacts={'scaler': 'scaler.pkl'})
   • Cmlflow.sklearn.log_model(scaler, artifact_path='scaler') and then log the custom model separately.
   • Dmlflow.log_artifact('scaler.pkl') after logging the model with mlflow.pyfunc.log_model().
   Reveal Solution Discussion Select an answer first
5. 4application · medium

   A team is deploying a model that needs to compute a rolling average of a numeric feature over the last 7 days. The raw input contains daily values, and the model was trained on the rolling average. They are building a custom PythonModel. Where should the rolling average computation be implemented?

   • AIn the training script, computing the rolling average and saving it as a feature column in the training data.
   • BIn the serving infrastructure, configuring a pre-processing step that computes the rolling average before the model is called.
   • CInside the predict method of the custom PythonModel, computing the rolling average from the raw input before passing it to the underlying model.
   • DIn the model's __init__ method, computing the rolling average for all possible inputs.
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A team is deploying a churn prediction model. The raw input features include a customer's signup date, and the model was trained on a feature called 'days_since_signup'. The team wants to package the model so that the deployed endpoint accepts the raw signup date and computes 'days_since_signup' inside the model's predict method. Which MLflow approach supports this?

   • ACreate a custom PythonModel class with a predict method that transforms the signup date to days_since_signup, then calls the underlying trained model; log it with mlflow.pyfunc.log_model().
   • BPrecompute days_since_signup in the training script and log it as a feature column in the training dataset, then serve the model on that precomputed data.
   • CLog the raw trained model with mlflow.sklearn.log_model() and rely on the serving infrastructure to apply the date transformation automatically.
   • DUse mlflow.models.Model() to register a custom model class without implementing the predict method, then call it directly from the endpoint.
   Reveal Solution Discussion Select an answer first