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

Notification

Icon
Error

84 Pages«<3031323334>»
Sepefeets Silverfish update (OG, Kara, MSG)
sepefeets
#629 Posted : Monday, July 18, 2016 11:46:36 PM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
sadpwner;39742 wrote:


http://pastebin.com/f8QWxCyV

Turn 1 should have coined into the egg but coined into flameimp?

__

Is it possible to lower the value of egg when it doesn't have any activators or isn't trying to play around aoe?

Data sync...

Probably too much thinking for an old card but maybe a small static penalty.
parakeet
#624 Posted : Tuesday, July 19, 2016 12:53:17 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 128

Thanks: 18 times
Was thanked: 83 time(s) in 69 post(s)
sadpwner;39706 wrote:
http://pastebin.com/S9FyqLvq

The bot missed lethal several times on turn 9. I assume it has something to do with the power word glory casted on the councilman.

Can someone tell me - if you are the dominant player, enemy casts power word glory on your minion, then you attack with it - does the minion damage trigger before or after the healing?

There is some datasync error on the extra power overwhelming. AI casts it, but the stats do not update, so it plays the sea giant instead. However, at several points, fatal is contingent on the minion damage triggering before the power word glory.

But regardless, bot should have done the random actions first. I think there is a bonus for doing random actions first, I am not sure why it wanted to attack first. It was a complicated board calculation and limited by calculation settings - that could be the cause.
1 user thanked parakeet for this useful post.
sepefeets on 7/19/2016(UTC)
parakeet
#630 Posted : Tuesday, July 19, 2016 12:54:29 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 128

Thanks: 18 times
Was thanked: 83 time(s) in 69 post(s)
@Sepefeets - Does ai know if it is the dominant player or not? Do we have access to that data?
1 user thanked parakeet for this useful post.
sepefeets on 7/19/2016(UTC)
sepefeets
#631 Posted : Tuesday, July 19, 2016 1:39:00 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
parakeet;39753 wrote:
@Sepefeets - Does ai know if it is the dominant player or not? Do we have access to that data?

I'm not sure what you mean?

I'm pretty sure power word glory triggers before damage is done but it needs to be tested to be sure.
cert
#633 Posted : Tuesday, July 19, 2016 1:40:19 AM(UTC)
Rank: Member

Groups: Registered
Posts: 19

Thanks: 5 times
Was thanked: 16 time(s) in 11 post(s)
Bug Report: it never use mark of y'shaarj whether there is a beast or not, or there is remained mana or not, in whole game
sepefeets
#634 Posted : Tuesday, July 19, 2016 1:48:04 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
cert;39756 wrote:
Bug Report: it never use mark of y'shaarj whether there is a beast or not, or there is remained mana or not, in whole game

Give a log.
sadpwner
#635 Posted : Tuesday, July 19, 2016 3:26:27 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 262
Location: somewhere

Thanks: 3 times
Was thanked: 64 time(s) in 54 post(s)
Is it true that the bot doesn't simulate the cards the opponent will play and only stimulates the attacks the opponents will do?

http://pastebin.com/0bkMaRuk

Should play councilman+villager
1 user thanked sadpwner for this useful post.
sepefeets on 7/19/2016(UTC)
parakeet
#637 Posted : Tuesday, July 19, 2016 3:38:56 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 128

Thanks: 18 times
Was thanked: 83 time(s) in 69 post(s)
sadpwner;39762 wrote:
Is it true that the bot doesn't simulate the cards the opponent will play and only stimulates the attacks the opponents will do?

This is mostly true. AOE Playaround is a simplistic simulation of what might be played. The AI also simulates casting a basic minion with power and toughness based on the available mana.
1 user thanked parakeet for this useful post.
sepefeets on 7/19/2016(UTC)
sepefeets
#636 Posted : Tuesday, July 19, 2016 3:40:25 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
sadpwner;39762 wrote:
Is it true that the bot doesn't simulate the cards the opponent will play and only stimulates the attacks the opponents will do?

For the most part yes. It is very dumb yet also huge portion of processing because for every one of your own potential boards it has to sim enemyTurnMaxWide (40 default) more enemy boards, although there is a big improvement coming...
parakeet
#632 Posted : Tuesday, July 19, 2016 3:51:02 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 128

Thanks: 18 times
Was thanked: 83 time(s) in 69 post(s)
sepefeets;39755 wrote:
I'm not sure what you mean?

I'm pretty sure power word glory triggers before damage is done but it needs to be tested to be sure.

Dominant player example: https://www.reddit.com/r...he_dominant_player_bug/

Yes, thanks, according to the advanced rulebook wiki, glory should trigger before the damage is dealt regardless of details.

1 user thanked parakeet for this useful post.
sepefeets on 7/19/2016(UTC)
sepefeets
#638 Posted : Tuesday, July 19, 2016 5:37:33 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
@sadpwner - Improvements for reliquary seeker and nerubian egg are done but fyi your enemy turn settings are ridiculous. I'm baffled that you don't rope every turn.

@parakeet - Interesting, I've never heard of that.

I'm not sure if you saw my comment on the pull request but it needs a little reworking.
FingerSkater
#640 Posted : Tuesday, July 19, 2016 8:33:49 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 45

Thanks: 14 times
Was thanked: 18 time(s) in 13 post(s)
V122.5SE changelog?
1 user thanked FingerSkater for this useful post.
sepefeets on 7/19/2016(UTC)
sepefeets
#642 Posted : Tuesday, July 19, 2016 8:34:40 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
V122.5SE released.

small penalty for dream
Mortal Coil penalty - factor in spellpower
fix _combo.txt switching checking for _mulligan.txt existence
563.1 reduce bonus on enemy card burn
cut duplicate enemy playfields too (from HB) + some cleanup (cutting playfields is a noticeable speed gain but don't view it as a reason to raise enemy turn calculations, they are very dumb, use the gains for your own turn)
625.6 (+1 old log) drastically reduce nerubian egg early bonus
(625.2, 625.3) avoid playing reliquary seeker without 6 minions ((625.1) intentionally not fixed, use cardvalue to fully prevent early play)
(pm) play brann before good battlecries
(491) play frothing berserker before attacking
penalize deadly shot for each small enemy minion



I'm regretting not bumping the major version after writing out the log...
5 users thanked sepefeets for this useful post.
Photonious on 7/19/2016(UTC), Dre on 7/19/2016(UTC), oijkn on 7/19/2016(UTC), parakeet on 7/19/2016(UTC), er009 on 7/19/2016(UTC)
sepefeets
#641 Posted : Tuesday, July 19, 2016 8:35:30 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
FingerSkater;39773 wrote:
V122.5SE changelog?

You're too quick, jesus!
parakeet
#639 Posted : Tuesday, July 19, 2016 2:03:02 PM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 128

Thanks: 18 times
Was thanked: 83 time(s) in 69 post(s)
sepefeets;39772 wrote:
@sadpwner - Improvements for reliquary seeker and nerubian egg are done but fyi your enemy turn settings are ridiculous. I'm baffled that you don't rope every turn.

@parakeet - Interesting, I've never heard of that.

I'm not sure if you saw my comment on the pull request but it needs a little reworking.

I did, checking a new settings.txt file for each each enemy class seems like a good idea since it offers more flexibility than behave=control:warrior:rush:pala:mana;... Just have not had the time to dig into it.
1 user thanked parakeet for this useful post.
sepefeets on 7/19/2016(UTC)
Dre
#645 Posted : Tuesday, July 19, 2016 3:01:00 PM(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)
@sepe, we once (long time ago) learned that in settings as a thumbrule this applies:

twotsamount at 20% of maxwide
enemyTurnMaxWide & enemySecondTurnMaxWide at 10% of enemyTurnMaxWideSecondTime
nextTurnTotalBoards at 20% of twotsamount
nextTurnMaxWide at 10% of nextTurnTotalBoards

the values of which the settings-generator is buildupon. (link to specfic forum)

Do these rules still apply you would say?
1 user thanked Dre for this useful post.
sepefeets on 7/19/2016(UTC)
sepefeets
#646 Posted : Tuesday, July 19, 2016 7:40:41 PM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 899

Thanks: 1028 times
Was thanked: 584 time(s) in 238 post(s)
enemySecondTurnMaxWide actually isn't even used for anything, only 1 enemy turn is sim'd. As for everything else I can't really say since those numbers are from empirical testing so you need to do empirical testing again. At least now you can see the action list easily at the end to compare. All I can really say is that the enemy turn sim is extremely dumb (look in enemyturnsimulator.cs, you should be able to tell most of what it's doing) but making it smart would kill performance.
sadpwner
#647 Posted : Tuesday, July 19, 2016 9:53:03 PM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 262
Location: somewhere

Thanks: 3 times
Was thanked: 64 time(s) in 54 post(s)
I rarely rope. It only happens when the board requires 4+ recalcs.

I didn't devote numbers to the maxwide because I couldn't get it to stress my cpu enough. I probably did something wrong before. What do you suppose the upper limit for maxwide is (the maximum amount of calculations theoretically possible)?

Btw, can you explain this line to me?

nextTurnMaxWide = 600; //maximum boards calculated in one second-turn-"combo-step"

__

Is there a repository that contains the misplays for 625.6, 491 and ect?
__

http://pastebin.com/9VppBVen

On turn 4, it chose murloc tidecaller from peddler despite me setting a much highest bonus to flame imp?

__

I spoke about it before but I forgot to go over my full idea.

It was about creating library of cards for each class. It will initially use a general library of cards we specify until it spots certain cards that we also specify. For example, spot flameimp = zoo deck (similarly to what track o bot does, the source code is available for that, maybe it's usuable) and it plays on the basis that the opponent has these cards. For each deck, we can specify a play behavior (rush, face, control) and also prioritise the secrets to play around e.g. explosive>freezing. In addition, you can create a library for your own deck and it will play with consideration of the cards in your deck.
1 user thanked sadpwner for this useful post.
sepefeets on 7/20/2016(UTC)
parakeet
#650 Posted : Wednesday, July 20, 2016 1:34:41 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 128

Thanks: 18 times
Was thanked: 83 time(s) in 69 post(s)
sadpwner;39786 wrote:
I rarely rope. It only happens when the board requires 4+ recalcs.

I didn't devote numbers to the maxwide because I couldn't get it to stress my cpu enough. I probably did something wrong before. What do you suppose the upper limit for maxwide is (the maximum amount of calculations theoretically possible)?

Btw, can you explain this line to me?

nextTurnMaxWide = 600; //maximum boards calculated in one second-turn-"combo-step"

maxwide can be huge depending on the board. With 7+ minions on the board there is often hundreds of thousands of possible ways to play the turn.

IIRC nextTurnMaxWide is how many boards will be simulated on your next turn. So the AI simulates the first turn, then enemy turn, then your next turn. Calculating next turns adds value, but not enough to be roping because the AI is trying to optimize your next turn. I would boost first turn simulations if I was looking for somewhere to spend extra computation time.
1 user thanked parakeet for this useful post.
sepefeets on 7/20/2016(UTC)
parakeet
#648 Posted : Wednesday, July 20, 2016 1:53:21 AM(UTC)

Rank: Advanced Member

Groups: Registered
Posts: 128

Thanks: 18 times
Was thanked: 83 time(s) in 69 post(s)
sadpwner;39786 wrote:
I spoke about it before but I forgot to go over my full idea.

It was about creating library of cards for each class. It will initially use a general library of cards we specify until it spots certain cards that we also specify. For example, spot flameimp = zoo deck (similarly to what track o bot does, the source code is available for that, maybe it's usuable) and it plays on the basis that the opponent has these cards. For each deck, we can specify a play behavior (rush, face, control) and also prioritise the secrets to play around e.g. explosive>freezing. In addition, you can create a library for your own deck and it will play with consideration of the cards in your deck.

Sepefeets has better simulation of your own deck on his to do list. Enemy class based behavior will likely be in the next update. Enemy deck based behavior is a bigger job and requires continual updating. If you demonstrate the value added it might be more likely to happen.
1 user thanked parakeet for this useful post.
sepefeets on 7/20/2016(UTC)
84 Pages«<3031323334>»
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 0.200 seconds.