Release Notes - 2019.09.25

General Improvements

Improvements to Data Pipeline Processing

Data Ingestion Bugs Fixed

OneAi Bugs Fixed

Expanded the logging of dropped columns in the EDA report.  (ref 479)

Relaxed row validation to allow single row predict frames. This feature allows users to get a single prediction back and is required to conduct certain types of time series regressions.  (ref 491)

One Ai : Forecaster. The forecaster allows users to configure autoregression to project value into the future. This feature is still in Beta so feel free to reach out to the OneModel team if you're interested in using this exciting new capability. (ref 499)

Implement CV Folds. User can now configure between a train test split or a cv fold approach to training and evaluating models. CV folds can lead to better, more stable evaluations of performance. (ref 498)

Support Quarters for Time Series Regression ( and other types ). We’ve expanded the types of sample dates that can be used. Previously, a user had to send a narrow spectrum of date time formats that excluded formats like quarters. Now with the assistance of the OneAi team, user can use almost any type of date model that OneModel supports. (ref 531)

Added time series train test splits. We’ve added a new problem_type of time_series_regression into the advanced section of OneAi. The new problem_type will tell OneAi to use a time series train test split regiment.  (ref 532)

Integrate Commute Time Augmentation. This feature is still in beta and will need to be turned on by the OneModel team. If you're interested in using this feature reach out to the OneModel team.  (ref 490)

Added specific error messages for errors with an SFTP data destination instead of a generic error message. For example, instead of showing “An error occurred running the data destination” for any error, an incorrect password will now return “Permission denied (password)”.   (ref 2575)

Improved the pipeline processing script validation to check all selected fields as part of a set of a GROUP BY.  (ref 2736)

Pipeline processing scripts now support the use of dynamic values when using the LEAD and LAG functions..  (ref 2809)

Provide the user with a specific error when configuring a delimited file from SFTP or as a Data Source fails due to missing "escape character" and "ignore quotes" fields. (ref 2824)

We fixed a bug where in rare situations the data consolidation service would continue to run if one of the sub-tasks failed, which could lead to a partially incomplete data load.  Now the service will only run and merge data when all tasks have successfully completed.. (ref 2810)

Fixed a bug causing value counts in modeling report to display incorrectly. This ended up being a problem with determinism in certain sklearn estimators. (ref 272)

Fixed a bug causing length mismatch errors to incorrectly display between training and testing sets. (ref 416)

 Fixed a bug with persisting feature selections derived from mutual information ( this manifested as a Lime error ). (ref 427)

Fixed a bug where OneAi would display a KeyError if two similarly named columns appears in different dataframes. (ref 442)

Fixed a bug that caused an error if a user didn’t specify a feature selection technique in the PCI section of the advanced configuration. (ref 489)

Fixed a bug that caused the suspicious column check to incorrectly show a zero point difference. (ref 502)

Fixed a bug that caused descriptive statistics to be empty on the EDA report while using a problem type of regression. (ref 521)

Fixed a bug causing the cheating column check to fail while conducting a regression. (ref 536)

Fixed a bug causing a KeyError when a label occurs in the predict dataframe that doesn’t show up in the training dataframes. (ref 552)