search

crowdAI / marLo

Multi Agent Reinforcement Learning using MalmÖ
MIT License
244 stars 46 forks source link

Implement Soft Reset #57

Closed ludns closed 5 years ago

ludns commented 5 years ago

According to this Malmo issue here you can implement soft reset in the mission.xml to make the training process faster (most of the time is taken by the reloading between two missions). I'm ok with implementing this but first I would like to know from @spMohanty and the other maintainers how best it would fit in the framework

spMohanty commented 5 years ago

@justinglibert : Thats a really good idea ! Given its an addition to the mission.xml, it can be easily added as an extra param in the game params, and then rendered conditionally using the jinja2 templating system we use to render the final mission.xmls. And then use the soft-reset conditionally during the env.resets

Please feel free to send in a pull request ! This would definitely help a lot of participants !

ludns commented 5 years ago

Just realised it's actually built into Marlo. Just set force_reset to false in the parameters dictionary when you init the environment.