Hi guys,
I didn’t use ml-agents before so this will be my first usage in a project, I want to create a game which will use the ml agent as an opponent for player in varying difficulties let’s say 100 different levels of difficulties but I want to have single brain with varying intelligence levels, I have checked but couldn’t find any info about it… is it possible or you have to have 100 different brain for it, which doesn’t make sense I hope there’s a sensible way for it… also I want to be sure that it’s totally okay to use it for mobile both ios and android because these are my target platforms… I read it runs for both platform but just in case I want to ask to be sure about performance and so on… any useful info is highly appreciated…
Hi @Neohun - a good first step is to try to get training on one type of level to make sure it works (i.e. implementation of rewards, etc…). After that, there are def different approaches in terms of having varying levels of intelligence. You could go with different brains (perhaps with different rewards or fewer fidelity observations). We’ve seen developers take this approach and swap models. Additionally, are currently looking at better ways to parameterized skills and active learning (we will have a blog post on this soon).
Also, we include an inference engine (Barracuda) that allows you to run the ml-agents models on iOS and Android. I would still suggest you do your own performance benchmarks but the barracuda package is intended to be performant.
Hope this helps.
Hi Jeffrey,
Thanks for giving a time and reply, so for now the only approach is using different brains for different levels, And I look forward to the approach you have mentioned above… it’s obviously the thing what I need but at least for now I have a concept in my mind to vary the difficulty such as decreasing the time limit for player and so on…
Regards,