Release Notes - 2020.04.22

Welcome to the 2020.04.22 product release.  This article provides an overview of the product innovations and improvements to be delivered on 22 April, 2020.  The article is structured as follows:

User Experience

One Ai Innovations

Performance & Platform Improvements

  • Performance Improvements

  • Data Ingestion Bugs Fixed and Minor Improvements

  • Improvements to Data Pipeline Processing

  • Minor Improvements and Bugs Fixed

  • One Ai Bugs Fixed

User Experience

  • When using the drill-through feature in a Storyboard the details returned in the results would include all the data from the base metric  In this release we have more effectively focused these results to include the filters applied to that metric from the UI (i.e. Explore or Storyboard Filters).  This encompasses dimension filters.  This should provide a much improved user experience and also provide an ongoing efficiency benefit for creating and managing your metric catalog requiring the creation of less base metrics and relying on UI filtering. (ref 3203)

  • In this release we added to Storyboards a feature missing in the original transition from the traditional Dashboards, which was the ability to copy a table to a data destination.  This feature is available from the table menu. (ref 4413)

One AI Innovations

  • Analytics with ethics is an important cornerstone of One Model’s product vision.  This drives our philosophy for delivering an open infrastructure and maximum traceability and explainability across the whole suite.  In this release we have built the foundations for a powerful overt Bias Detection capability as part of One AI. (ref 771)

  • In this first release we are delivering a focus on core group fairness metrics and configuration such as:
    -    The ability to exclude protected classes from being trained upon.
    -    Providing specific EDA metrics, such as, disparate impact and mean difference from pre-predicted data.

  • In upcoming releases, we will expand the number of pre-modeling measures of group fairness as well as introduce post prediction fairness reporting and de-biasing strategies.

  • In this release we introduced the concept of wrapper feature selection methods. Wrapper methods allow One AI to use recursive feature elimination strategy with various algorithms as the source estimator. We’ve also released a hybrid method that allows One AI to reduce the overall feature space using blended filter methods ( depending on the variable type ) to reduce the feature space and then a wrapper method to ultimately decide which and how many features to use. Both One AI and users can utilize these methods to build robust feature selection pipelines and ultimately improve model performance and EDA. (ref 752)

  • Results Summary Chart Descriptions - details about how to interpret the charts on the result summary have been added. (ref 742)

Performance & Platform Improvements

Performance Improvements

  • We delivered some notable performance improvements to the One Model platform in this release, with up to 15% improvement in performance of queries metric queries. This difference is specific to queries that include a Dimension (including Filtering and RBS) - and the larger and more numerous the Dimensions, the greater the improvement should be. 

  • For the technically minded, this was done by loading the entire Dimension structure into each node of the Redshift cluster (known as DISTSTYLE ALL). This means, instead of having to load any missing parts of the Dimension into the node as required for the query, the entire structure is already there before the query is run.   

  • We wanted to mention this in the release notes because performance has such a big part to play in the user experience and let you know it’s an area of continual innovation. (ref 4213)

Data Ingestion Bugs Fixed and Minor Improvements

  • Fixed an issue where tasks in a cancelled/resumed API might not run if they didn't start before the cancel. (ref 4518)

Improvements to Data Pipeline Processing

  • Fixed an issue where metrics with either the "Same Period Last Year", or "Previous Period" Time Function, and the "To Date" flag enabled would not return correct results when the support for Pay Periods was turned on.  Note: This "To Date" flag is used to restrict the period referenced to the most recent run date.. (ref 4407)

  • In this release we have delivered some performance improvements when editing Storyboards.  Particularly when you are editing a Storyboard with a large number of charts/tables etc. and changing the layout, modifying text tiles, titles etc. there was noticeable UI lag.  After this release the experience should be a lot snappier. (ref 4249)

  • We fixed an issue that could cause a Storyboard to crash if a user that doesn’t have permission to a particular dimension that is in the Storyboard Filter Template.   The Storyboard will now open, but not display the restricted dimension.  If the user has edit rights to the Storyboard they will be able to see the restricted dimension in the Filter Template even though they won’t be able to open it and see any of the nodes for security reasons. (ref 4309)

One AI Bugs Fixed

  • Fixed the Prophet Forecaster Logistic Capacities  (ref 800)

  • Fixed an issue with ADASYN Upsampling (ref 801)


Was this article helpful?

0 out of 0 found this helpful



Please sign in to leave a comment.