[Rock Paper Scissors]Any way to beat abbey in a more certain way?

Python
#1

Tell us what’s happening:

I am trying to build a strategy in [Machine Learning Rock Paper Scissors] project.
I am using Q-learning with greedy strategy.
I found I always beat abbey about 50%～60% percent, which is a not good result. The other three players are more easily to beat. Are there any way to beat abbey in a more certain way? Thanks.

Your code so far
https://github.com/cmal/rock-paper-scissors/blob/master/RPS.py#L50-L74

Your browser information:

User Agent is: Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/85.0.4183.102 Safari/537.36.

Challenge: Rock Paper Scissors

Link to the challenge:

#2

I think abbey’s pattern changes over time. I used a response table that I build each time through based on the opponents play history and player’s play history but I only look at the last 100 plays. That seemed to get me to a win rate of about 67% for abbey. When I use the entire history I only get about 55% win rate with abbey.

#3