• Kraken leverages the latest scikit-learn permutation importance to calculate Driver Influence for your binary, multi-class and regression models, using the automatic holdout data from your input dataset. Driver Influence is calculated independently for each algorithm in your analysis. All new analyses will use the new Driver Influence calculations automatically; you can also retrain your existing models and compare new Driver Influence values to those calculated previously. Click here to get all the details on Driver Influence.

• We've updated two hyperparameters used in Random Forest algorithms to better accommodate a wide variety of input datasets, resulting in improved predictive accuracy and less chance of overfitting. All new models will use the new hyperparameters automatically; you can retrain existing models to generate a new version of your analysis that uses the new hyperparameters. 
  • The number of estimators is 100 (previously it was 10)
  • max_depth is 10 (previously it was infinite)

• When adding a Data Pipeline "Find and Replace" step on string fields, you have two new options for more flexible and robust find and replace actions:
  • Allow partial matches (to search on part of a string)
  • Replace text with empty string (to remove text)

Reminder: SONAR© Guide content is just a click away!