Univariate drift detection for hotel booking dataset
In the previous exercises, we established using the multivariate drift detection method that the shift in data in January is responsible for the alert in the ROC AUC metric and the negative business value of the model.
In this exercise, you will use a univariate drift detection method to find the feature and explanation behind the drift.
The reference and analysis sets are already pre-loaded for you.
Deze oefening maakt deel uit van de cursus
Monitoring Machine Learning in Python
Oefeninstructies
- Specify Wasserstein and Jensen-Shannon method for continuous methods and L-inifity and Chi2 for categorical.
- Fit the reference and calculate results on the analysis set.
- Plot the results.
Praktische interactieve oefening
Probeer deze oefening eens door deze voorbeeldcode in te vullen.
# Intialize the univariate drift calculator
uv_calc = nannyml.UnivariateDriftCalculator(
column_names=feature_column_names,
timestamp_column_name='timestamp',
chunk_period='m',
continuous_methods=[____, ____],
categorical_methods=[____, ____],
)
# Plot the results
uv_calc.____(reference)
uv_results = uv_calc.____(analysis)
____.____().____()