I left the simulation running from Friday night. When I came back today, my computer had reset. On Friday I continue from save file 110610-1716. The format is year, month, day, hour, and minute. YYMMDD-HHMM.
The simulation created four subsequent save files.
- 110610-2144, 199 MB
- 110610-2251, 199 MB
- 110610-2358, 199 MB
- 110611-0104, 199 MB
It's strange that files are 199 MB when 110610-1716 was only 26.9 MB. If the population hit my population cap of 512 robots, the simulation should have paused and no further save files should have been made.
I'm also not sure whether the last save is valid or corrupted because of the crash. I'd love to know whether the crash was due to my code or a power failure.
At least the save files are less than 426 MB, 110609-2346.
At the moment I'm not happy with the rate of neural network updates. Movement and energy drain are based on time but the neural network updates are not which doesn't feel quite right.
I'm going to have to lower the population cap to about 256 robots because having 512 robots is too unwieldy to deal with.
I've also had a read of the Bullet Physics sample code and it looks like using constraints to connect blocks is relatively straight forward. The awesome part is that once connected, applying force to the constraints is relatively straightforward as well.
I think that the rate of energy drain of 1 unit per second is too low. I also think that the rate of food dropped into the world is too high.
I was fairly unfocused today. I spent a fair big of time working on "Games I'm Going to Play Damn It!".
Grr. Loading a 199 MB binary file takes ages as well. Hmm. There were 594 robots and 19 units of food in the 199 MB save file. Reducing from 594 to 297 to X robots.
Increasing energy drain from 1 unit per second to 2 units per second.
Grr. Building with serialization enabled takes a long time.
Hmm. I don't think when I changed the serialization to serialize variables I hadn't before and upped the version, that it worked properly.
RELATED
No comments:
Post a Comment