Certified Machine Learning Professional

Concept drift refers to a change in the relationship between input variables and target variables over time. This change can cause the model's predictions to become less accurate as the original relationship the model learned is no longer valid. It is not merely a shift in the distribution of input or target variables but specifically in how they relate to each other.

To assess if there is a statistically significant association between the time period and the presence of missing values in a particular value of a categorical input variable, the appropriate tool would be the Two-way Chi-squared Test. This test is designed to determine if there is an association between two categorical variables – in this case, the time period (old vs. new data) and the presence or absence of values within a specific category. The one-way Chi-squared test is generally used for goodness-of-fit tests and is not suitable for assessing associations between two categorical variables.

In order to nest the child runs under the parent run, the child run blocks need to be indented within the parent run block. This way, the child runs are executed within the context of the parent run, allowing MLflow to correctly nest the runs.

A machine learning engineer who wants to log feature importance data from a CSV file within an MLflow run should use the function mlflow.log_artifact. This function is designed to log files or directories as artifacts of a run, allowing for easy tracking and analysis along with other relevant model information. Therefore, the correct code block to use is mlflow.log_artifact(importance_path, 'feature-importance.csv').

A simple, low-cost method of monitoring numeric feature drift is to track summary statistics trends. This involves monitoring basic statistical metrics such as mean, median, standard deviation, and other relevant summary statistics over time to detect any shifts in the distribution of the feature values. This approach is straightforward and does not require complex statistical tests, making it both simple and cost-effective.