Closed Jamesflynn1 closed 1 year ago
24/10: Read through Game Theory-Based Opponent Modeling in Large Imperfect Information Games (first pass excl. full experimental results). Created Github project and issues to track progress and provide stepping stones in the right direction.
25/10 Read part of Morrhill and von Stengel and wrote their entries into the GitHub issue
26/10 Read part of intro to CFR again, researched what Poker game to play primarily looking at work by Jeary. Decided on 2 player Limit texas hold 'em for all initial work. Defined future work and created the milestone for the progress report.
28/10 Supervisor meeting, added more issues to Github
Tabula: "Discussed which game to focus on, abstraction methods, opponent exploitation and potentially using EFR vs CFR and what benefits that might bring."
31/10 Read part of Evaluating State-Space Abstractions in Extensive-Form Games . Reading Hindsight and Sequential Rationality of Correlated Play to help understand EFR. 1/11 Continued reading Hindsight and Sequential Rationality of Correlated Play and started reading the follow on paper.
2/11 Supervisor meeting : "Discussed EFR, I will write on the motivation and intuition for the algorithm. Project will likely look at how opponent exploitation can be applied along such an algorithm." Started with EFR writeup
3/11 Continued with EFR writeup. Found that deviations permitted by EFR results in a payoff increase when playing against a more static opponent (data from Morhill's EFR paper). Would be interesting to see performance against exploitable opponents.
6/10 Continued with EFR writeup in Overleaf.
07/11 Understanding line by line the EFR algorithm. Continued with EFR writeup, added EFR algorithm, wrote notes on EFR. Developed more intuition regarding correlated equilibria. Need to further develop understanding of CFR by using https://www.ma.imperial.ac.uk/~dturaev/neller-lanctot.pdf.
08/11 Reading CFR intro thoroughly to help to writeup
09/11 Started progress report added literature review section and project management section.
Use this thread to keep track of progress (and save it into a log maybe).