PredictionResult

Introduction

Backend Name: prediction_result
Description: If your game is using any predictive models (e.g. Multi-arm Bandit) or other machine learning algorithms to pull levers in your game, use this event to track the prediction related to the player. Since players may have different values, this is used during analysis to track model predictions to behavior outcomes.
Event Type: Experiments
Priority: P1

Parameters

Name	Type	Select	Backend name	Description
additionalData	Dictionary	Optional	additional_data	Any additional information
modelInput	object	Required	model_input	JSON formatted data with model’s input. The actual input can be key-value pairs or arrays depending on the model.
modelName	string	Required	model_name	The name of the model which could be descriptive to its usage.
modelOutput	object	Required	model_output	JSON formatted result from calling the model. Either single key-value pair or multiple key-value pairs.
modelVersion	string	Required	model_version	The version of the model

Usage

// TODO Rephrase (@Bhaumik Shukla)

Model details served by service. This event is track to the inputs, outputs and related information by calling a model and returned by the server. It starts tracking when the game calls the model, and fires the event when the game receives the result. If the call takes over X seconds and not getting response, we should fire this event anyway to let us know the call is failed.

Code example:

LionAnalytics.PredictionResult("ml_model_trial_25", "0.1.25", new int[3] { 5, 4, 6 }, new int[3] { 4, 4, 4 });