Katsuki Ohto

Results 26 comments of Katsuki Ohto

feature: apply lambda=1 in the timestep that there is no value output

VTrace looks working well in experiments with Geister.

Adding multidiscrete feature

Hi, @Jogima-cyber ! Long time no see, and thanks for your great suggestion! I also tried the multiple action situation this year, so I'll check if it can be put...

Adding multidiscrete feature

@Jogima-cyber I pushed **feature/multi_unit** branch into my fork. https://github.com/DeNA/HandyRL/compare/master...YuriCat:HandyRL:feature/multi_unit In this branch, each agent can output `env.num_units()` actions in each turn. I've confirmed that it works with TicTacToe!

Adding multidiscrete feature

@Jogima-cyber > unit_mask Indeed! We need `unit_mask` if there are absent units.

Adding multidiscrete feature

@Jogima-cyber `.unsqueeze(-1)` may be necessary for either of the two.

Adding multidiscrete feature

@Jogima-cyber Updated my branch [feature/numti_unit](https://github.com/DeNA/HandyRL/compare/master...YuriCat:HandyRL:feature/multi_unit).