Roger Larsson

I have no need of `rng` in `observation`, completely deterministic, and calls to `split` function to generate new are among the slowest in my loop. state - global state (`transform`...

In **my** project `observation` is completely deterministic - playing Ludo. Unless you want to restrict usage of Gymnasium to simulating physical systems. I only need randomness when `transforming` state(board), action...

1. If something is added to an API before 1.0 you will have to live with it almost forever... 2. Sadly the Perfetto trace coverage stopped working for me. Done...

I see no reason for `terminal` to be noisy. Having `key` in all methods will require generating a new one before **all** calls, and `split` takes time... My thought was...

But `to_vec` also define a vector... The vector did not exist before this method is called.

Prior to release of 1.0 would be the perfect time to remove this - no API promises on alfa/beta... If someone needs backward compatibility they can use old git version...

I do something like this ` for p_, info, trajectories in Game.play_tournament(_a_gym_env, agents, record_trajectories=True, games_to_play=BATCH_GAMES, seed=seed): ` where `play_tournament(...) - - - p_observation, _info = a_gym_env.reset(seed=seed) - - - while...

HumanRenderer.close() does both ``` pygame.display.quit() pygame.quit() ``` so by closing the environment pygame should be terminated correctly - there should be no need for a direct call to `pygame.display.quit()` If...

There is a point in letting the HumanRender be transparent, as it actually is not possible to observe/verify what is rendered. The other modes could be streamed to file or...

Can this be closed now?