Andrew Kemendo comments

Results 7 comments of


                                            Andrew Kemendo

Cannot train with multiple instances due to grpc errors

Thank you for the issue. We will look into it and get back with you soon, thanks!

Assets\ML-Agents\Examples\PushBlockWithInput\Scripts\PushBlockActions.The type or namespace name 'InputAction' could not be found (are you missing a using directive or an assembly reference?)

Thanks for the bug report. We’ve documented your input and will come back to you once we determine how best to support your request.

Is it better to use self.step_queue.get() than self.step_queue.get_nowait() in SubprocessEnvManager._step()?

Thank you @Tyushang we will log this request and get back with you once we have investigated it

Add advices for training looking for tensorboard

Thanks for your questions @mercurion. Your cumulative reward graph isn't showing the hallmarks of catastrophic forgetting however it doesn't seem to be converging and your reward variance seems to be...

Make (most) internals internal protected

Hi Benedict, Thanks for the suggestion. We’ve documented your feedback and will come back to you once we determine how best to support your request.

Make (most) internals internal protected

Thank you @RedTachyon we'll add your input to the ticket as a +1 to the feature request.

Workload imbalance of clients(or SubprocessManager step not evenly over all envs).

Thank you @Tyushang we will log this request and get back with you once we have investigated it