find trainer parameters that work well (evolution leads to strong AIs)

• send click from UI to board
• show the state of the board on UI
• show dice
• make possible to make only one cross as an active player
• prevent UI from closing if invalid turn -> show e.g. error on UI instead
• make possible to cross as a passive player
• show valid crosses directly after clicking
• add skip button (for active and passive player)
• show if player is active or passive
• make UI smaller (max. 1/4 of screen)

-Problem:
luck, other players, order of players and strength of players have an influence on the current fitness metric

• Solution:
  get strength of an AI by playing several times, in different constellations; add points/placements

compare to #4

• complete a row only if allowed
• take extra points for completing a row into account

• visualize progress
• let one AI play multiple times -> How do results change? How do they reflect the strength?
• standard deviation of player's points -> look at Gaussian distribution (narrow/ wide), How many games for average? Test for different AIs with different settings and use it to justify the metric. #4
• evaluate human players for comparison to AIs -> show why AI performs better
• execute trainer with same settings multiple times -> Does it result in similar AIs? -> reliable or random algorithm? -> which hyperparameters are important?
• Does it make sense to only accept stronger generations? (assumably not)
• How many parameters are optimal for the strategy? (ax^n+ bx^(n-1)…)
  • Find typical critical situations (e.g., one cross in the red row with last number 5 , two crosses in yellow row with last number 5 -> where to cross the 6?) => only with a quadratic function you know that you should cross yellow
• Reduce strategy vectors from 9 to 5 elements (no actual strategic difference between red/ yellow ; green/ blue), at least one strategy for the opponents

Hello! I also have written an AI to play Qwixx in Python! (Matthews-Makes/Qaike)

I think it would be really nice not to have to run the commands that you have listed.
You could definitely set up Gitpod very easily.

Gitpod.io is the website

sum_situation = cx²+bx+a (before: bx + a)