notes.txt

pistonball24

./run_optimize.sh [name] # run on each to get parameters in DB

./run_best_hyperparameters.sh [name] # generate jsons

./run_eval[n].sh # save mature policies and print hyperparameters with best performance

./run_render[n].sh # generate saved gifs

./run_render_optimization.sh # generates gifs from policeis logged during optimization if something weird happens


Todo:
Fix rendering logic


For new runs:
Change DB name
Change mature saving treshhold
Change env in evaluation, rendering and training files
Change neural networks (bird flocking)
Change number of steps

Weekly match: top left, with table on left and other shit on right. Only left thing scrolls


Leaderboard: 2x4 page of top 8 games that show you the match against player 1 and 2 of that game last week as gifs and link to specific leaderboard pages



OMP_NUM_THREADS=1 python3 train.py --algo ppo --env LunarLanderContinuous-v2 -n 2000000 -optimize --n-trials 150 --sampler tpe --pruner median --study-name pistonball23 --storage mysql://root:dummy@10.128.0.28/pistonball23



OMP_NUM_THREADS=1 python3 train.py --algo ppo --env LunarLanderContinuous-v2 -n 2000000 -optimize --n-trials 150 --sampler tpe --pruner median --study-name pistonball23 --storage mysql://root:dummy@10.128.0.28/pistonball23