Returns the saved predictions from a validation of a model. This function can be used to access historical predictions in ranking system performances, screen backtests, and strategy simulations. It allows for much longer simulations, with lower execution time, than those using AIFactor(). However, certain restrictions apply (see below), and you must enable “Save Predictions” before doing the cross validation of the model.

Parameters

"AI Factor Name": The name of the AI Factor. 
“Model Name”: The name of the validation model. 
“Dup_id”: Optional in case there are multiple copies of the same model

How to use

When starting the cross validation of an AI model, enable the option to "Save Predictions". When the validation is complete, click the fx icon next to the model's name, which should be enabled. This will pop up a dialog with additional usage information including the pre-filled AIFactorValidation() formula. 

NOTE: If you forgot to enable “Save Predictions”, you will need to delete the model (or create a duplicate) and redo the validation. 

Usage restrictions

In order to use stored validation predictions, the following is necessary:

1. The universe must match
2. The dates used in the backtest must exist in the AI Factor's validation holdout data.
3. The rebalance period must match

Examples

1) In a Ranking system, simply add a Stock Formula and use the AIFactorValidation() function as the Formula. You can use multiple AIFactorValidation() in a ranking system or combine AIFactorValidation() with other typical ranking system rules.

2) In a Screen rule to get the top decile:

FRank(`AIFactorValidation("My AI Factor", "lightgbm II")`,#All, #Desc) > 90

3) In a Screen rule to get the bottom decile:

FRank(`AIFactorValidation("My AI Factor", "lightgbm II")`,#All, #Desc, #NANeutral) < 10