FIX: PREDICT function does not return error message for missing columns or handle learning parameter for Tree/Forest models in SQL Server 2017 ML Services

Applies to: SQL Server 2017 Developer on WindowsSQL Server 2017 Enterprise on WindowsSQL Server 2017 Enterprise Core on Windows

Symptoms


Assume that you use the PREDICT or sp_rxPredict function in SQL Server 2017 ML Services in the following scenarios:

  • When you use the PREDICT function to generate predicated values or scores based on a rxDTree, rxDForest, or rxBTrees model, and the columns that are needed to make the prediction are missing from the input data of the WITH clause, the execution would fail without a detailed error message that states which column is missing.
  • Learning rate in the tree models (rxDtree, rxBtrees, and rxDforest) was not included in the serialized real-time model. Therefore, a default value (0.1) instead of the user-specified value was being used after unserialization. This affects both real-time scoring and native scoring when you use the sp_rxPredict and PREDICT functions.

Resolution


This issue is fixed in the following cumulative update for SQL Server:

       Cumulative update 4 for SQL Server 2017

Status


Microsoft has confirmed that this is a problem in the Microsoft products that are listed in the "Applies to" section.

References


Learn about the terminology that Microsoft uses to describe software updates.