by Peter » Wed Dec 07, 2016 10:02 pm
Hi Gary,
my own agent I'm trying to develop.
I use reinforcement learning and it seems to be independent of the level.
However, it mostly happens the next time it zooms in after an exploration action. (I wanted to write only but it just happend after an exploitation action)
There are random exploration actions and calculated exploitation actions, neither does interact with the server.
The agent zooms in, takes a screenshot, calculates the remaining number of birds, zooms out, chooses an action, execute that actions (and repeat)
And the server output is identical for both.
Server output for one shot:
Client 421337 Message: doScreenShot
Client 421337 Message: doScreenShot
Client 421337 Message: getState
Client 421337 Message: doScreenShot
Client 421337 Message: fullyZoomIn
Client 421337 Message: doScreenShot
Client 421337 Message: fullyZoomOut
Client 421337 Message: doScreenShot
Client 421337 Message: doScreenShot
Client 421337 Message: pFastshoot
Shoot from: (156 323 ), offset/polar: (-4 48) at time 0
Shooting Completed