Welcome to the latest One Model product release update. This article provides an overview of the product innovations and improvements to be delivered on 18 May 2022. You can see the full 2022 release schedule here: https://help.onemodel.co/en/articles/3134972-one-model-product-release-calendar.
Over the last few months we have been working on a major upgrade to our development framework upgrading to the latest version of .net. This upgrade provides lots of improvements in our base development technology and sets us up for some great innovations in the future. As we mentioned in the previous release notes, we have modified our release schedule a little to deploy this new platform update. Given this change, we are going to skip the release planned for 25 May 2022.
While we have performed extensive testing, please let us know if you see anything unusual and we will jump right on it!
In addition, we have lots of very cool enhancements to our One AI capability and you can read more about these below.
One AI Innovations
New Regressors: Nine new regression algorithms added to predictive regression problem types. These algorithms were always being used in the backend of One AI but now we have GUIs that allow users to customize settings and exclude specific regressors.
Probability calibration in the frontend: This feature allows you to smooth out the probabilities that come out of One AI so that they will play nicer with buckets (high, medium, low). We use this setting for our standard storyboard for attrition ( more on that below ).
New Standard Storyboard for Attrition Models: We’ve done a lot of work to help users build great content with One AI – the most recent effort has been around attrition models. We’ve built some standard content to help users build driver analysis storyboards. If you’re interested in this let us know and we can set it up for you.
Faster SHAP Values for LightGBM: We Updated SHAP value calculations for lightGBM ( an algorithm that One AI uses ). This feature allows SHAP values to be calculated many times faster than they were previously – giving the user robust understanding of models with less time.
New Result Download Options: We Updated the files that can be downloaded from the Results Summary Reports on the Augmentations Page – we’ve added more files and file metadata that will cause the files to open in excel by default.
Improvements to Commute Time: We’ve added a distance attribute in the results of our commute time augmentation so users can see how far employees are from the office. We’ve also added a new commute time value that doesn’t take into account traffic. The existing commute time that does factor in traffic periodically returns empty values when there are road closures and obstructions. This issue doesn’t manifest when traffic is excluded from our calculations. In cases where a user is calculating commute time for long distances they may want to use the non-traffic value as it will be more stable.
Improved Feature Selection: In this release we’ve added a few enhancements to the feature selection processes within One AI – these changes will result in One AI making feature selections that are more stable, resulting in better overall model performance.