Release Notes - 2019.08.28

Added a New Data Destination for Amazon S3

One Ai Innovations

Improvements to Data Pipeline Processing

 Data Ingestion Bugs Fixed

One Ai Bugs Fixed

Customers are now able to use Amazon S3 as a data destination in addition to the SFTP destination that was previously available. This allows data from Metrics and Dimensions, Augmentations, and One Model Data Sources to be exported on a regular basis to a Customer's S3 Bucket. (ref 2212) 

For details on how to use Amazon S3 for Data Destinations, refer to http://help.onemodel.co/en/articles/3208494-configure-a-data-destination 

The One Ai team added advanced configurations for CV folds, giving users control over the number of folds via an advanced configuration key called cv_params.  The cv_params key accepts a dictionary containing kwargs to be passed to the cross_validation method during processing i.e. {'number_of_folds': 5}.  (ref 198)

New regression types have been added into the advanced configuration section of One Ai: SVR, Ridge, HuberRegressor, GaussianProcessRegressor, RandomForestRegressor, SGDRegressor, LogisticRegression. (ref 482)

Per Column Type Coercion:  In the advanced configuration section of One Ai users can now manually set the type (i.e. continuous or categorical ) of columns. This feature is useful if One Ai's auto column classification makes a mistake or if you want to ensure that a column gets processed as a specific type.. (ref 404)

Improved the stability of data loads, in some cases where there are network issues within AWS. (ref 2177)

If one of the tasks for the SF API connector for One Model fails, the error message presented in the Data loads status page read "Complete with Errors" - this will now read "Failed". (ref 2257)

Fixed a bug where a Data Load would not be given an Errored status if it errored while warming Cache (ref 2431)

In the Loads Status page One Model updated the following API load statuses:“Not Run” to “Waiting to Start”“Cancelled as Already Active” to “Rejected - Previous load still running”. (ref 2512)

Manual Selecting Columns displayed the incorrect number of columns previously. (ref 432)

Some columns being dropped via the correlation checks weren't actually getting dropped. (ref 479)

Fixed a value error caused by column name mismatching across data frames. (ref 442)

Using sklearns RandomForest estimator can occasionally create non-deterministic results, which was leading to validation failures. (ref 272)

Fixed a bug that dropped columns that were set to be undroppable via PCI settings. (ref 522)

Fixed a bug that prevented persisted models from storing mutual_info feature selection results correctly. (ref 427)