Judgement Task Implementation

In todays lecture we implemented the basics of the judgement task. Here is the code for everyone to download:

Unpack those files and replace your files in the project with these new ones and all should be good.

Cheers Arend

One Comment

  1. Jory Schossau March 25, 2014 11:47 pm Reply

    We followed Arend’s example in class today and I had a bug so worked on my code a bit. Then I tried a couple things that I didn’t get to in class and it seemed interesting so I thought I’d share. Please share too if you do something you think is an interesting approach regardless if it works well or not (like evolution!). Although this is just a sample problem, we can share some different types of approaches to playing with the fitness evaluation.

    The following approach got me consistently between 90% and 100% correct answers from the agents. Initially I tried just punishment of -0.5 for wrong answers of the two answers we were scoring. While that did fairly well it seemed inconsistent between replicates and it had difficulty getting above 80% accuracy. So in addition to punishment I added an interpretation of a third output to indicate “no decision” for which I awarded 0.005 (other values were attempted but this worked best). The idea was to encourage delayed judgement.

Leave a Reply