Adding support for population-based training for variable-length reinforce-based training. Credits to @eugene-kharitonov
Note that signature and methods for variable-length reinforce-based training do not change. Current games using SenderReceiverRnnReinforce are not impacted.
Motivation and Context
Training and supporting multiple listeners and multiple receiver can be useful for population-based (possibly generation-based) games.
How Has This Been Tested?
UTs with new population file pass (note: these includes games using the SenderReceiverRnnReinforce class)
Description
Adding support for population-based training for variable-length reinforce-based training. Credits to @eugene-kharitonov
Note that signature and methods for variable-length reinforce-based training do not change. Current games using
SenderReceiverRnnReinforce
are not impacted.Motivation and Context
Training and supporting multiple listeners and multiple receiver can be useful for population-based (possibly generation-based) games.
How Has This Been Tested?
UTs with new population file pass (note: these includes games using the SenderReceiverRnnReinforce class)