|
@startuml serial_collector |
|
header Serial Pipeline |
|
title Serial Collector |
|
|
|
|#99CCCC|serial_controller| |
|
|#99CCFF|env_manager| |
|
|#CCCCFF|policy| |
|
|#FFCCCC|collector| |
|
|
|
|#99CCCC|serial_controller| |
|
start |
|
:init collector, set its \nenv_manager and \ncollect_mode policy; |
|
|#99CCFF|env_manager| |
|
repeat |
|
|#99CCFF|env_manager| |
|
:return current obs; |
|
|#CCCCFF|policy| |
|
:<b>[network]</b> forward with obs; |
|
|#99CCFF|env_manager| |
|
:env step with action; |
|
|#CCCCFF|policy| |
|
:process transition; |
|
|#FFCCCC|collector| |
|
:save transition in cache; |
|
if (for every env: \n env_i is done? <b>OR</b> cache is full?) then (yes) |
|
if (is <b>sample_collector</b>?) then (yes) |
|
note right: Only sample_collector will do so, \n episode_collector will not. |
|
|#CCCCFF|policy| |
|
:<b>[adder]</b> get train_sample from cache; |
|
endif |
|
|#FFCCCC|collector| |
|
:save sample/episode for return; |
|
if (env_i is done?) then (yes) |
|
|#99CCFF|env_manager| |
|
:env_i reset; |
|
endif |
|
endif |
|
|#FFCCCC|collector| |
|
repeat while (collected sample/episode is not enough?) |
|
:return sample/episode; |
|
stop |
|
|
|
@enduml |
|
|
