How to Customize the Machine Learning Property Files

In the Machine Learning Files page, you can download one or more of the properties files, modify them in a text editor and upload them back to the service.

When customizing the machine learning properties files, you must keep the following guidelines in mind:

  • Include 2-3 examples of high-quality responses.

  • Ensure examples cover different scenarios relevant to your use case.

  • If possible, try to use close to real data with sensitive information anonymized.

  • Ensure no sensitive information is provided in the prompt.

  • NOTE: You must not rename the downloaded file. Uploaded file must match the downloaded file name.

Example file format

Example1:

Data:

<<Dimension/Headers/Keys>>

<<Members/data_values/Values>>

or

<<Dimension/Headers/Keys>>

<<Members/data_values/Values>>

<<Calculations>>

or

<<Dimension/Headers/Keys>>

<<Members/data_values/Values>>

<<Top Contributors data>>

Response:

<<Expected Narrative>>

Note:

  • If any of the Machine Learning properties files have been customized, the new default properties files need to be modified to include the previously made changes.

  • There will be a notification with a warning icon next to the properties file in the Machine Learning files window.

Example for Compare

Data1:

[Account, Entity, Period, Years, Scenario, Property, Currency, Country, Version, Union Code, Component, value]

[EquityPerShare, Discover Bank, Qtr4, 2001, Fcst vs Bud, OWP_CYTD(Prior), EUR, Britain, V1, Total Union, Total Component, 1032.6]

[EquityPerShare, Discover Bank, Qtr3, 2001, Fcst vs Bud, OWP_CYTD(Prior), EUR, Britain, V1, Total Union, Total Component, 987.4]

Change in amount = 45.2

% Change = 4.58%

Response:

In Qtr4 2001, the Fcst vs Bud EquityPerShare for Discover Bank, Britain was EUR 1032.6 for Total Union and Total Component of OWP_CYTD(Prior) Property. It was EUR 987.4 in Qtr3, indicating an increase of EUR 45.2 or 4.58%.

Using Comments in the ML Template Files

You can add comments to the ML template files by enclosing the comments in /* and */, such as:

/* This is a comment on the example */

Translating Gen AI Responses into Different Languages

You can have the Gen AI textual responses translated into different languages by updating the ML template files. At the bottom of each template file, you can change the desired language in the commented section:

/* Language instruction (optional): Use the exact same verbatim. Respond in {your preferred language}, e.g. "Respond in French" */

Language instruction: Respond in English

Note:

  • The translation is generated directly by the Large Language Model rather than by Reports. As a result, any translation issues, inconsistencies, or variations cannot be modified or corrected.
  • The translation is performed directly on the generated text, dimension and member names/aliases are not treated any differently than the surrounding text, from a translation standpoint.
  • The modified ML template file for language selection is selected by the report designer in the Criteria Definition for a report grid, which applies to all users accessing the report. Different language translations cannot be presented based on a user’s locale.