search

alex-petrenko / sample-factory

High throughput synchronous and asynchronous reinforcement learning
https://samplefactory.dev
MIT License
773 stars 106 forks source link

Minimize memorizing? #282

Open jarlva opened 10 months ago

jarlva commented 10 months ago

Hello,

Using a gym env problem with time-series data the model quickly memorizes. I'm looking for ways to minimize it by injecting noise and dropping frames by inserting noise/dropping layers in the model (note link below). Is there a simple way to do that? An example is more than welcome.

https://github.com/alex-petrenko/sample-factory/issues/260#issuecomment-1383485703

alex-petrenko commented 10 months ago

Did you try to look at data augmentation? Is there any way to synthetically generate data in your domain, or synthetically generate more samples from the ones you have?