Welcome Guest! To enable all features please Login or Register.

Notification

Icon
Error
84 Pages«<2728293031>»
Sepefeets Silverfish update (OG, Kara, MSG)
Options
Previous Topic Next Topic
parakeet
#564 Posted : Tuesday, July 12, 2016 6:50:34 PM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 128

Thanks: 18 times
Was thanked: 83 time(s) in 69 post(s)
sepefeets;39543 wrote:
That tiny penalty doesn't explain why excessive multi thousand penalties to playfield value don't make it pick a board that doesn't attack doomsayer, or why it takes excessive bonuses to get it to select another board. It's a dead end anyways because changing playfield value only affects which one is chosen, but if all the boards have bad action lists then it can only pick from bad to worse. In order to have "do nothing" as an option, the actions have to be denied from ever being added to at least some of the playfields because of the penalty manager or move generator.


We have to not add value for minions on the board - see my pull request. Sometimes the second turn calculation makes those 1k board value penalties not matter - as is the case with using a weapon to attack into doomsayer. By then the doomsayer is already gone but the weapon bonus is still there. Dre's #439.2 is an example of this second turn (turncounter = 1) effect. EDIT: and the getPlayfieldValue retvals are 8 for the correct non-attack move and 9 for the incorrect attack move. With penalties of -1 and -5 respectively (i think).

For those who are looking into it:
The value of "do nothing" actions is calculated. It goes through each possible board one move at a time and calculates the boardvalues. First is 'no action', then 'only one action', then 'only two actions', up until all possible moves have been exhausted or calculation limits are reached. Even if the playfield is not complete, meaning the possible moves have not been exhausted, if the board value is higher, it will keep it.
The deep value is how many moves that playfield entails
The len is how many playfields were simulated
The dones is how many of those playfields have no more possible moves.

You can see it always ends calculations with the len = dones as long as calculation limits have not been reached. This implies the possible actions have been exhausted. And yes of course, it can sometimes miss moves.

So for example, playing a weapon and not attacking with it is simulated as a possible board.
Back to top
WWW
 |  Edit by user
1 user thanked parakeet for this useful post.
sepefeets on 7/13/2016(UTC)
sepefeets
#569 Posted : Tuesday, July 12, 2016 8:57:03 PM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
You're right, I hadn't looked much at mini simulator but I see it starts with 0 action playfields. Something still feels terribly wrong about playfield value not responding to huge changes when a normal playfield value is only a few hundred.
Back to top
WWW
parakeet
#570 Posted : Wednesday, July 13, 2016 1:37:07 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 128

Thanks: 18 times
Was thanked: 83 time(s) in 69 post(s)
sepefeets;39553 wrote:
You're right, I hadn't looked much at mini simulator but I see it starts with 0 action playfields. Something still feels terribly wrong about playfield value not responding to huge changes when a normal playfield value is only a few hundred.

Yeah, that was bothering me also. No matter what I put for the boardvalue I could not make it pick the better move, +1,000 or -1,000, it made no difference, the Eaglehorn Bow always targeted doomsayer. But when I stepped through, I saw that I was only making the turncounter=0 board look better, while the weapon bonus made the following, turncounter=1, turn look better. The bot then picks the misplay because it thinks the board position will be better after the opponents turn.

The nice thing about the playfield value approach is that these types of things do not need to be individually coded:

Back to top
WWW
1 user thanked parakeet for this useful post.
sepefeets on 7/13/2016(UTC)
sepefeets
#571 Posted : Wednesday, July 13, 2016 6:33:11 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
160 has something to do with secret playaround giving 50 bonus for playing the doomguard to trigger possible snipe and another 70 to trigger possible explosive trap but there is only 1 secret so it should only get one of the bonuses, but I can't trace where it's coming from. See if you can find it please.

543 kills the doomsayer when I sim it with your fix.

Good work.
Back to top
WWW
sepefeets
#572 Posted : Wednesday, July 13, 2016 7:13:45 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
V122.3SE released.

Some updates from HB 281 / inspired by it (cards that cause discarding, vol jin, forbidden ritual).
552.2 (no log) - penalize mortal strike above 12 hp.
Updated to support the new shaman hero.
Reworked playing around doomsayer.
Back to top
WWW
1 user thanked sepefeets for this useful post.
er009 on 7/13/2016(UTC)
Dre
#573 Posted : Wednesday, July 13, 2016 9:43:19 AM(UTC)

Rank: Advanced Member

Groups: HSModerator, Registered
Posts: 888
Man
Location: The Netherlands

Thanks: 216 times
Was thanked: 480 time(s) in 315 post(s)
Still from the logs of 122.2SE (day before yesterday)

http://pastebin.com/EqURCw5X

Uses houndmaster on a spider on turn 6 - arguably not bad, because it could attack right away altough the elek was a bigger target - but then on turn 7 it plays another houndmaster and uses it on the same spider which had taunt allready. Just to attack face. I dont see where this was the better play instead of taunting another minion up like the elek. in this case. There was no 5 health treat on the board.
Back to top
WWW
1 user thanked Dre for this useful post.
sepefeets on 7/13/2016(UTC)
oijkn
#575 Posted : Wednesday, July 13, 2016 11:00:24 AM(UTC)
oijkn


Rank: Advanced Member

Groups: HSModerator, Registered
Posts: 472
Man
Location: 127.0.0.1

Thanks: 76 times
Was thanked: 271 time(s) in 93 post(s)
Hi Sepefeets,

Thank you for your great work and contribution ! I hope you will continue to update this great project :)

For people who are interested i updated the my SFU tool, now it's pointing to Sepefeets's GitHub.

Link to SFU Topic

Best regards.

PS : If you don't agree with it just notify me and i will delete it !
Back to top
WWW
2 users thanked oijkn for this useful post.
sepefeets on 7/13/2016(UTC), jsmwood729 on 7/14/2016(UTC)
sadpwner
#576 Posted : Wednesday, July 13, 2016 11:12:26 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 262
Location: somewhere

Thanks: 3 times
Was thanked: 64 time(s) in 54 post(s)
I noticed an error with the new silverfish.



oijkn;39572 wrote:
Hi Sepefeets,

Thank you for your great work and contribution ! I hope you will continue to update this great project :)

For people who are interested i updated the my SFU tool, now it's pointing to Sepefeets's GitHub.

Link to SFU Topic

Best regards.

PS : If you don't agree with it just notify me and i will delete it !



If you have time, do you mind updating the class switcher to include _discovery.txt
Back to top
WWW
 |  Edit by user
1 user thanked sadpwner for this useful post.
sepefeets on 7/13/2016(UTC)
oijkn
#577 Posted : Wednesday, July 13, 2016 11:18:17 AM(UTC)
oijkn


Rank: Advanced Member

Groups: HSModerator, Registered
Posts: 472
Man
Location: 127.0.0.1

Thanks: 76 times
Was thanked: 271 time(s) in 93 post(s)
Hi sadpwner,

Ok well i will try to do it, i don't want to spam/topic off this thread.

Regards.
Back to top
WWW
1 user thanked oijkn for this useful post.
sepefeets on 7/13/2016(UTC)
er009
#578 Posted : Wednesday, July 13, 2016 1:58:33 PM(UTC)
er009


Rank: Advanced Member

Groups: Registered
Posts: 238
Man
Location: EU

Thanks: 61 times
Was thanked: 56 time(s) in 42 post(s)
dragonpriest crash log, as reported earlier that dragonpriest tends to randomly crash silver no matter what.

http://pastebin.com/ALVFV55j

Back to top
WWW
1 user thanked er009 for this useful post.
sepefeets on 7/13/2016(UTC)
etronpoilu
#580 Posted : Wednesday, July 13, 2016 4:54:26 PM(UTC)
etronpoilu

Rank: Member

Groups: Registered
Posts: 26
Location: Can

Thanks: 17 times
Was thanked: 13 time(s) in 11 post(s)
Just installed V122.3SE and got those errors when it reads carddb.txt :



Those cards are specific for the Mechazod Tavern brawl so shouldn't be a big problem.. Is it only me who get this ?
Back to top
WWW
1 user thanked etronpoilu for this useful post.
sepefeets on 7/13/2016(UTC)
sepefeets
#581 Posted : Wednesday, July 13, 2016 5:51:22 PM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
@everyone - Don't worry about the unidentified tavern brawl cards.

@er009 - Okay that log triggers the crash, I'll get it fixed.
Back to top
WWW
er009
#582 Posted : Wednesday, July 13, 2016 8:01:27 PM(UTC)
er009


Rank: Advanced Member

Groups: Registered
Posts: 238
Man
Location: EU

Thanks: 61 times
Was thanked: 56 time(s) in 42 post(s)
Shaman Brawl crash - http://pastebin.com/EcdQhZSE
Back to top
WWW
1 user thanked er009 for this useful post.
sepefeets on 7/13/2016(UTC)
parakeet
#579 Posted : Wednesday, July 13, 2016 9:19:08 PM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 128

Thanks: 18 times
Was thanked: 83 time(s) in 69 post(s)
er009;39589 wrote:
dragonpriest crash log, as reported earlier that dragonpriest tends to randomly crash silver no matter what.

http://pastebin.com/ALVFV55j


For now, you might be able to avoid the crash by taking Deathwing, Dragonlord out of the deck.
Back to top
WWW
1 user thanked parakeet for this useful post.
sepefeets on 7/13/2016(UTC)
parakeet
#583 Posted : Wednesday, July 13, 2016 9:26:25 PM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 128

Thanks: 18 times
Was thanked: 83 time(s) in 69 post(s)
er009;39600 wrote:
Shaman Brawl crash - http://pastebin.com/EcdQhZSE


This can be fixed. For now, it should be avoidable by not playing with mimironshead.
Back to top
WWW
1 user thanked parakeet for this useful post.
sepefeets on 7/13/2016(UTC)
parakeet
#574 Posted : Wednesday, July 13, 2016 9:53:49 PM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 128

Thanks: 18 times
Was thanked: 83 time(s) in 69 post(s)
Dre;39567 wrote:
Still from the logs of 122.2SE (day before yesterday)

http://pastebin.com/EqURCw5X

Uses houndmaster on a spider on turn 6 - arguably not bad, because it could attack right away altough the elek was a bigger target - but then on turn 7 it plays another houndmaster and uses it on the same spider which had taunt allready. Just to attack face. I dont see where this was the better play instead of taunting another minion up like the elek. in this case. There was no 5 health treat on the board.

This kind of misplay makes sense to me because the boardvalue calculation places no value on taunt for your minions. Said differently, giving one of your minions taunt does not increase your board position according the boardvalue calculation. I have been hoping someone would produce a misplay related to this so I would have an excuse to add value to taunt. However, if I add value for taunt, it does not get rid of the misplay - so something else must be going on. D:
Back to top
WWW
1 user thanked parakeet for this useful post.
sepefeets on 7/13/2016(UTC)
sepefeets
#584 Posted : Wednesday, July 13, 2016 10:14:06 PM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
A new build with fixes for the crashes will be out soon.
Back to top
WWW
1 user thanked sepefeets for this useful post.
Svenek on 7/14/2016(UTC)
sepefeets
#585 Posted : Thursday, July 14, 2016 12:23:38 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
V122.4SE released.

573 - fix deathwing dragonlord crash (HB 281)
552.1.2 - avoid replacing weapons with weapons from spells/battlecries
578 - fix mimiron crash
add missing card id's
Back to top
WWW
1 user thanked sepefeets for this useful post.
jimbob70 on 7/14/2016(UTC)
sadpwner
#586 Posted : Thursday, July 14, 2016 1:51:07 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 262
Location: somewhere

Thanks: 3 times
Was thanked: 64 time(s) in 54 post(s)
Can't we add a profile for arena and just make it a long-term goal? You only need 4 wins from arena to profit for 20-30 gold on average. 5 wins is the ideal goal for the bot netting you 50-60 gold profit. That's half of what we farm everyday, which is a lot.
Back to top
WWW
1 user thanked sadpwner for this useful post.
sepefeets on 7/14/2016(UTC)
Dre
#591 Posted : Thursday, July 14, 2016 3:17:40 AM(UTC)

Rank: Advanced Member

Groups: HSModerator, Registered
Posts: 888
Man
Location: The Netherlands

Thanks: 216 times
Was thanked: 480 time(s) in 315 post(s)
Logs from last night version V122.3SE - Mid-range Hunter (currently rank 7 - 55% Wr)

http://pastebin.com/ptTv5mYa edit: @joy had to put this in [url] to make it clickable

Turn 2, bot uses here power when it has a 2-drop in hand. I think this is the penalty for flamejuggler when it doesnt have a chance to hit a minion. This value (3 i tough) imho should be 0 (not even 1) because it's only a chance not a fact. and it still puts pressure on the board even when there are no minions, and allways atleast hits face for 1. Maybe there should be only a penalty for this when the enemy has a enrage minion on the board as only minion so you have a 50% chance to enrage it.
Back to top
WWW
 |  Edit by user
1 user thanked Dre for this useful post.
sepefeets on 7/14/2016(UTC)
84 Pages«<2728293031>»
Forum Jump  
You cannot post new topics in this forum.
You cannot reply to topics in this forum.
You cannot delete your posts in this forum.
You cannot edit your posts in this forum.
You cannot create polls in this forum.
You cannot vote in polls in this forum.

Powered by YAF | YAF © 2003-2011, Yet Another Forum.NET
This page was generated in 0.158 seconds.