new python AI, needs testing! Current version: 0.4.7

Discussion of all aspects of the game engine, including development of new and existing features.

Moderator: Forum Moderators

Sir Lamorak
Posts: 8
Joined: February 26th, 2008, 7:08 pm
Location: Camelot

Post by Sir Lamorak »

I just played your ai and I think its pretty good. I did notice a couple odd things about it though.
First, all the ai units seem to run away with little provocation and with little sense of strategic placement: i.e. a horse on full health running away to a stream where he can be trapped and killed easily.
Second, related to the first point, although the ai is good about grabbing towns it seems unconcerned about actually holding onto them.
Finally, the king moves oddly. He moves every turn regardless of whether he needs to or not and he spends a good deal of time away from his castle. Also, he has the same problem as the other units: he runs away/gives up towns without batting an eyelash and he generally ends up in an exposed location. I won the game as quickly as I did largely due to the fact that the king ran from his town into a pond right when my level two drake was coming in.
I hope those pointers help.
The slithey toves did gyre and gimble in the wabe.
yobbo
Art Contributor
Posts: 151
Joined: September 16th, 2005, 6:31 am
Location: New Zealand

Post by yobbo »

Thanks for your feedback :).

I'll work on all the points you mentioned.

I've been somewhat stuck on the best way to make units brave, but I think I've cracked it.
SKAcz
Posts: 72
Joined: March 14th, 2005, 11:50 am
Location: Czech Republic
Contact:

Re: new python AI, needs testing! Current version: 0.4.7

Post by SKAcz »

Hi,
I tried tommy-0.4.7.py on 5p_Auction-X.map,
1x local player and 4x tommy AI, and i didnt recruit (only made end of turn) and in turn 3 always wesnoth falling down.
When I am using for example skaczian004.py, no problem is. I guess You have to use something special in python code. I have 1.7 GHz CPU with 1GB memory.
mesilliac
Developer
Posts: 81
Joined: November 18th, 2005, 8:09 am
Location: New Zealand

Re: new python AI, needs testing! Current version: 0.4.7

Post by mesilliac »

I tried to reproduce the problem using wesnoth 1.4 but I couldn't :(. I let the AI fight until turn 9 or so, when they killed my leader. I didn't notice any problems. I don't suppose you can run it from a terminal and tell me what the output is?
User avatar
ylmson
Posts: 15
Joined: March 24th, 2008, 10:08 pm

Re: new python AI, needs testing! Current version: 0.4.7

Post by ylmson »

Very interesting AI. Here's an extremely short reply on random map, default AI (red) vs tommy.py (blue), to illustrate a problem:

The notable problem in the replay is when the blue leader got attacked in turn 2, nobody came to save him. (And in the first turn the blue leader probably made a worse mistake by choosing to move to an exposed village without bringing enough protection with him)
Attachments
Random_map_replay.gz
Random map replay with Defualt AI (upper left) vs. tommy.py 0.4.7 (bottom right)
(6.77 KiB) Downloaded 262 times
"You may draw, even if you are no artist, and all you have is mspaint and a touchpad."
Post Reply