search

harukaki / brl

reinforcement learning for bridge
Apache License 2.0
8 stars 1 forks source link

Check speed #12

Closed harukaki closed 9 months ago

harukaki commented 1 year ago

PPOの学習速度を調べる

harukaki commented 1 year ago

new_ppo.pyの実行速度 image

harukaki commented 1 year ago

200 minで250 step 1 stepあたり、4,096 * 64 = 262,144 steps 1 stepあたり、116,805 board 1 stepあたり、50 sec 1 secあたり、5,242 steps, 2,336 boards

harukaki commented 1 year ago

image 山口さんの実行速度

harukaki commented 1 year ago

1 stepあたり、1000 board ((11 24 + 8) 60 + 20) 60 secあたり、135 1000 * 1000 boards 980,000 secあたり、135,000,000 board 1 secあたり、137 boards