SimpleRSAModelUttKLDivParamABK_iterative: Cost function for three parameter optimization for the...

In haniaelkersh/rsa-publish-test: Prior Inference Package

Description

Simple RSA

Softness, non-obedience and klValueFactor are optimized.

Usage

SimpleRSAModelUttKLDivParamABK_iterative(params, data)

Arguments

params

Three value vector specifying three of the three parameters to be optimized. softPrefValue: A parameter value between [0,infinity) (The larger the value the higher the tendency towards uniform liking). Value reflects how categorical the listener's preferences are: 0: The listener always picks her preferred object. If the listener prefers red objects, she will always pick the red object in the scene. infinity: It is as likely for the listener to pick green, blue or red objects. non-obedience:This parameter determines the extent to which the instruction of the speaker is obeyed by the listener. (0 = full obedience, infinity = full instruction ignorance). Example: 0: Listener always picks red objects following the utterance "red". infinity: Listener as likely to pick green, blue or red objects even if the utterance is "red". klValueFactor: A parameter that can be negative, 0 or positive: zero Don't care about learning about feature preferences of the listener positive Care about learning about feature preferences of the listener negative Trying to pick non-ambiguous utterances

data

A matrix with data rows. column structure: [1:OC1,OC2,OC3,4:numUttOptions,6: picked utterance] 1:OC1 Object 1. A value between 1 and 27. 2:OC2 Object 2. A value between 1 and 27. 3:OC3 Object 3. A value between 1 and 27. 4:numUttOptions The number of valid utterances in the scene. 6: picked utterance The utterance picked by the participant.

Details

This is the iterative version of SimpleRSAModelUttKLDivParamABK.

This function uses SimpleRSAModelUttKLDiv_3params_iterative.

Value

Minimized Kullback-Leibler divergence and the optimal parameters.

haniaelkersh/rsa-publish-test documentation built on Jan. 31, 2021, 2:02 a.m.