syzygy wrote:I would think giving the opponent more time is, for many reasons, far more practical than giving the opponent more cores.
But equally inefficient.
When testing Houdini I don't want to give 80% or 90% of the CPU resources to the other engine. It's not a good use of processing power if the engine you're actually trying to develop only gets a small fraction of the available CPU power...
syzygy wrote:I would think giving the opponent more time is, for many reasons, far more practical than giving the opponent more cores.
But equally inefficient.
When testing Houdini I don't want to give 80% or 90% of the CPU resources to the other engine. It's not a good use of processing power if the engine you're actually trying to develop only gets a small fraction of the available CPU power...
Well, it should at least be less inefficient.
Of course you should try to be as efficient as possible, so in general testing against inferior opponents is a bad idea. But if for some reason self-play is not helping anymore and all other engines are clearly inferior, then you need to do something
But the idea might be more useful for developers of weaker engines. They might get more out of testing against Houdini with 10% of the CPU resources than out of self-play or testing against Houdini with 50% of the CPU resources.
syzygy wrote:I would think giving the opponent more time is, for many reasons, far more practical than giving the opponent more cores.
But equally inefficient.
When testing Houdini I don't want to give 80% or 90% of the CPU resources to the other engine. It's not a good use of processing power if the engine you're actually trying to develop only gets a small fraction of the available CPU power...
Well, it should at least be less inefficient.
Of course you should try to be as efficient as possible, so in general testing against inferior opponents is a bad idea. But if for some reason self-play is not helping anymore and all other engines are clearly inferior, then you need to do something
But the idea might be more useful for developers of weaker engines. They might get more out of testing against Houdini with 10% of the CPU resources than out of self-play or testing against Houdini with 50% of the CPU resources.
That is an idea that Don promoted and we use to some degree with Gaviota.