Release Notes - 2023.08.23

Welcome to the latest One Model product release update. This article provides an overview of the product innovations and improvements to be delivered on 23 August 2023. 

One AI

• One AI Machine Learning Model Selection Enhancements 

In this release the One Model team has deployed a number of enhancements that bring more control and transparency to the model selection process for machine learning.  When a machine learning augmentation is run in One AI, a number of algorithm and setting combinations are tested and the best performing combination is selected for the final model.  This process is referred to as model selection.  What follows are several enhancements to this process that are included in the current release. (ref 2392)

• • Situational Intelligence for Model Selection - Situational intelligence introduces the capability to determine the best settings based on the context and characteristics of the data. The primary advantages of this approach are a reduction in the time it takes to generate models and predictions as well as a reduction in overfitting for certain algorithms.  While this feature is currently in its early stages, One AI will begin to leverage this capability to improve speed and accuracy when executing Recipes.  The static problem statements of Recipes lend themselves well to this approach.

• • Model Scoring Criteria Customization - It is now possible to control which metric is used to score model performance for model selection in One AI. Prior to this release, all scoring was performed using the F1 weighted score.  This default can now be overridden with precision or recall.  Either precision (the proportion of positive identifications that were actually correct) or recall (the proportion of actual positives that were identified correctly) can be the more important measure depending on the problem the model is attempting to solve.  In a future release we plan to introduce selection or creation of custom algorithms for scoring here.  These settings can be configured in the YAML Template section of the augmentation configuration using the following key:

              score_type: precision_weighted / recall_weighted

              # f1_weighted is the default

• • Messages and Warnings  - To further extend transparency into what happens when a machine learning augmentation is run, we have introduced user facing Messages and Warnings.  The Message and Warnings section can be found at the bottom of the Results Summary report for pending or deployed machine learning augmentation runs.  It displays information about the models created during the model selection process as well as setting overrides applied during the configuration of the augmentation.  The target audience for these messages is primarily Data Scientists.  Although we plan to add additional messages and warnings incrementally over time, the current list includes the following: 

• • • Which algorithms were tested and scored for model selection?
    • For each algorithm, was it run due to default settings, a heuristic, or a user override?
    • Were there any setting combinations that were rejected due to the data fed to the model not meeting certain criteria?

User Experience Improvements

• Users page

We improved the Users page where Admins create and modify users! Previously, you would type a username into the search box then click on a menu item and be directed to the respective page but, if you clicked the back button, the Users page would reset and you would have to start all over again. Now, when you click the back button, the searched User page retains the information you have put into the search box along with your selections including; search terms, Hide Inactive status, page, and items per page. (ref 15176)

• Accessibility 

We have made improvements to the screen reader functionality for chart exports from a tile. Now, if a user navigates to a tile in a Storyboard and selects the Export Chart option, the chart name will match the tile name, and an alert is spoken twice before the export commences. (ref 15763)

Data Processing Improvements

• Data Destination scheduling

In this release we are adding a much requested new feature to the Data Destinations capability.  Data Destination Users can set a Data Destination schedule to run after model processing has been completed. This new feature adds a higher level of automation and resiliency by automatically running the Data Destination when your data has updated, which can vary from day to day based on how much data is being processed. (ref 4961)

• Workday connector

We have made some performance improvements to the Workday Connector for non-Worker tasks. These will be gradually rolled out and tested behind the scenes. (ref 15640)

Minor Improvements & Bugs Squashed

• We fixed a bug with the color palette in Explore that was adding multiple color selections and users were unable to delete some of the extra colors that appeared. Now when you add a color to your color palette in Explore, you get just the one color selection and, subject to your permissions, you can delete it too. (ref 16532)
• We fixed a file processing issue affecting the Asia/Pacific region where files over 500GB would error during the validation phase. We made a few adjustments so file processing can complete. (ref 16991)
• We fixed an issue that arose after the last release where Processing Scripts could take hours to update Datawarehouse Relationships in the Data Model. Each Relationship was taking 1.5 seconds to run, so customers with a large number of Datawarehouse Relationships could be stuck in processing for some time. We fixed the issue and processing scripts are back up and running smoothly. (ref 16985)
• We resolved a problem where updates to the data behind the Get_References endpoint in Workday could result in some references being missed or extracted twice if the changes occurred while the OneModel Workday Data Source was running. (ref 15725)
• We made a minor improvement to support a timestamp with more than 4 fractions of a second in a data source table. For example, we previously could only include a date value with the precision of 2023-08-23 00:00:00.0000, but can now support date values with the precision of 2023-08-23 00:00:00.0000000 (ref 16787)
• We fixed a bug on the Users page where the pagination would include both active and inactive users, even if you had filtered out inactive users. We applied a fix so now you will only see what you have chosen whether that is active users only or both active and inactive users.  (ref 14959)
• We fixed a bug where some users were receiving an error message when saving to an existing multi-page Storyboard. We made a few adjustments and now users with permission can save their changes as expected. (ref 17179)