Release PV650 SV221

The theme of this minor release is performance enhancements to Sensible Machine Learning.

Prerequisites

  • Use OneStream Software platform version 6.5.0 or later with this version. 

  • This solution must be installed in a controlled environment. Reach out to your OneStream Software representative for questions regarding the installation process.

New

None.

Changed

Sensible Machine Learning

  • Clearer time series visualization on Analysis Prediction phase of Utilization.

  • Actuals time series no longer display gaps when Forecast Range is set to a shorter time frame than Target Data Source update time frame.

XperiFlow Engine

  • The external data source for stock ticker prices has been swapped out for a new data source with greater data reliability.

  • Improved worker logic for prioritizing of tasks and jobs.

 

Fixed

Grouped Model Health Score Calculation

  • The healthscore distribution plot and overall healthscore value show a grouped healthscore metric as 0 instead of as the aggregation of all targets in the group.

Known Issues

Sensible Machine Learning

Consumption Groups Can Export Data from Failed Predictions

  • If a prediction job were to fail after predictions were generated, the predictions would still be exported in the consumption group exports depending on settings chosen in the consumption group.

Cannot Edit/Delete a Location That Is Assigned to a Source Feature

  • If a source feature dataset uses the location dimension, any locations used here cannot be deleted or edited unless the feature dataset is no longer in use.

Feature Generalization Source Feature Issues

  • Statistics in the Feature Generalization grid on the Pipeline > Features page and Insights > Features pages can be misleading for source features. All target collection lags and source feature collection lags are considered the same feature in the feature generalization calculation.

Two users importing event packages prior to any existing

  • If two users run the import all event package job at the same time from Solution Setup prior to the job ever being run for the environment, then the job can fail. It is recommended to have one user run this job on initial login to avoid the issue.

Failures can occur in-app if datasource is altered without datasource update

  • Performing any of the following actions on source data without running a data source update can result in application failures:

    • Removing tables

    • Changing table names

    • Changing or removing column names

    • Having duplicate date and target combinations

The forecast range column not representing change during manual rebuild

  • On the Utilization Phase Manage > Audit page, the Forecast Range column does not show changes during a manual rebuild. If you change the forecast range in between rebuilds, then the original forecast range does not display properly in the Manage > Audit page, which still shows the old forecast range setting but not the new one.

The Distribution Statistics plot and related statistics no data case

  • In the Explore Targets and Features dashboard, the Distribution Statistics bar chart and related statistics have no data if the feature time range does not overlap with the time range of the target data set.

Loss of multiselect parameters in all multiselect Combo-boxes

  • There are situations when using multi-select in Sensible Machine Learning where the selection and the actual value of the selection do not match up. This does not cause major issues but may cause confusion where you have made a specific selection but the actual selection is different.

Pipeline Run and Pipeline Features dashboards rendering timeout

  • Pages in the Pipeline section may time out. There is a OneStream platform issue with the way BI Viewer dashboard components display data during the Model Build phase. There are certain cases when navigating throughout the Pipeline > Run and Pipeline > Features pages where OneStream times out. You may need to navigate around the Pipeline section to continue the Model Build.

Source data while job running

  • You cannot change a target or feature data source while a job using that target or feature data source is running.

Source data while building

  • You cannot change a target or feature data source while a model build is in progress.

Restarted Model Build have no build info status information

  • In the Utilization Section of Sensible ML, there is no way to tell the status of a model build (whether it successfully completed or failed). This makes it tough for the user to know the status of their project - especially in the event of an auto rebuild failing which will automatically archive the failed build for you. When that happens, there is no easy way to tell whether the build failed or not other than looking at the job activity table or the OneStream Task Activity info.

XperiFlow Engine

Audit page shows generated features across all models

  • The Utilization phase Manage > Audit page shows the number of features generated across all models for each target, instead of just those generated for the best model.

Lagged Database Connectivity on Initial Environment Build

  • Sensible Machine Learning has an initial database connectivity lag when installing an AIS bolt-on environment. There is a maximum 24 hour period where the DataSense Framework does not appear as a database connection option. Your organization should perform an IIS restart on creation of AIS.

Incorrect Allowed Models in Advanced Settings Page for Grouped Targets

  • In the Modeling phase Configure > Model page, models selected to run for a grouped target using the advanced settings may not actually run. Some models selected are ignored by the engine. If only these models are selected, at least one allowed baseline still runs and produces predictions for the grouped target.

Tasks/Jobs not copied over for copied projects

  • A copied project does not create copies of the jobs or tasks that ran from the original project. This means that those prediction jobs do not appear on pages such as the Utilization phase Manage > Predict page.

Job reversion is currently shown in the app

  • Job reversion functionality in Sensible Machine Learning should only be used with help of support and under known circumstances.

Targets with no data in the validation, test, or holdout portions of the data splits

  • In the Modeling phase Pipeline > Train page, if you have a target with no data in the validation, test, or hold portion, the mean model with a perfect fit for cleaned data is chosen. This then leads to the model having zero (0) for the error metric score. This makes the model appear to be performing well. These results should be ignored.

Discrepancy Targets: Rebuild small targets with a lot of cleaned data

  • After rebuilding a small target with a lower number of data points than the largest target in a project, the small target:

    • Incorrectly shows an expected number data points on the Explore Targets and Features page based on the largest number of data points found across the project. This is not the amount making up the rebuild.

    • The Modeling phase Data > Dataset page incorrectly shows the number of unique dates based on the largest target found in the data set. This is not the largest target found in the rebuild.

Only utilizes numeric source features

  • The engine only uses numeric source features. Non-numeric source features can be fed into the engine but are not used by any models. Strings, booleans, and other features are currently not supported.

Deleting a build does not delete objects created during that build

  • When deleting a model build, previously created project-level objects are not deleted. This includes but is not limited to events, locations, and source feature data sources.

Contacting Support

OneStream Support can be contacted at any time by registering at http://support.onestreamsoftware.com and then emailing support@onestreamsoftware.com.