Question Manually filling SAC experience buffer

In short:

I wish to collect experiences from everything from artificial training data to direct engagements with real players. How could i input experience from multiple separate computers?

 

Thinking through it some, one possible solution is to have the user’s computer record an entire game’s worth of states, upload that to the server, then force the model to relive it.

The problem with that though is that it would only work if i had it be deterministic. The model would then have to be the kind that is “stochastic” (i probably misspelled that) which i would prefer it not to be..

 

Yet another solution is to have a game mode where you specifically play online against the currently live models in the server. In that case i would need to have networking working and this would probably have to be run on the server, in training mode.

That way everything would be handled on the server, except for the players and i would not have to pull anything from user’s computers.