Testing Stall cards

[godofdeath500]

I understand your idea on picking USEM as the no-brain beating deck, but even as simple deck as that will sometimes have some inconsistent starting hands in short tests. Since the trainer allows you to give "illegal" decks for AI, you should think using something like this for opponent:

by godofdeath500

Hover over cards for details, click for permalink

Deck import code : [Select]

7ac 7ae 7ac 7ae 7ac 7ae 7ac 7ae 7ac 7ae 7ac 7ae 7ac 7ae 7ac 7ae 7ac 7ae 7ac 7ae 7ac 7ae 7ac 7ae 7ac 7ae 7ac 7ae 7ac 7ae 8pn

I think he uses USEM as the deck that he uses to beat the AI. But I agree with you on the illegal decks option. It would make the test much easier, in my opinion.

Also, why not use the decks in the ttw study, to see how fast they are after they are subject to stall? It seems to me that using the most popular rush deck is the best option.

[Ekki]

Not only interesting, I guess this will be really useful if given some time... Maybe aplying some mathematic formula you could even go further and explain the game mathematically, but I'm going too far 
Whatever, I'll have to think of some of the main uses this could have, and I might help you with your measures after some weeks, once my RL stops "bothering" me

[Glitch]

Quick question:  What if the stall deck wins?

[ratcharmer]

Quick question:  What if the stall deck wins?

Well, that hasn't come up yet . . .

Generally the test decks are so low damage that they stand no chance o winning, but if one comes up that can beat the deck I'm running against it Then I'll treat win% the same way it's treated in the ttw studies.

Notably, this will cause some bias in the results since it would mean not counting games where the AI won, and hence not counting some of the games where the AI had the most success stalling.

If/when this comes up I'll provide win% in the notes section of whatever I was testing.

@godofdeath & zse:
I am the one playing USEM, but I suppose I could reverse the roles if need be. Generally I want to see how well cards work against real decks people use though. If I tested cards against a deck of all pillars & frogs then it would just be showing how effective the card was against frogs.

Some inconsistency is unavoidable, but when you're playing against the same opponent every time the results should be much more consistent than simply grinding AI3 etc, so I can get away with 20 games instead of 100.

I actually did pull USEM from the ttw study, and I am planning to test other decks from there. In general these are my criteria:

-Needs to be fast enough that I won't go crazy running 20 turn games
-Can't be so fast that the AI never gets to use the card I give it
-Needs a high win% and to be fairly consistent on draws.

[Ekki]

Quick question:  What if the stall deck wins?

Well, that hasn't come up yet . . .

Generally the test decks are so low damage that they stand no chance o winning, but if one comes up that can beat the deck I'm running against it Then I'll treat win% the same way it's treated in the ttw studies.

Notably, this will cause some bias in the results since it would mean not counting games where the AI won, and hence not counting some of the games where the AI had the most success stalling.

You could take them in account somehow. I mean, if you kept on playing against those matches where you lose, you'd have more stats, and if you don't, you'll avoid some information (relevant, since it would decrease the stall index rating).
You could also take in account a win%, and then make two ratings for both cases. That way you'd have a stall rating for % of win against that counter, and another stall rating taking in account both won and lost games (I believe only taking in account won games might be unuseful).

[jmdt]

Seeing how USEM essentially plays itself.  Using the 'stalldeck' may give more accurate results

[ratcharmer]

@Krathos: I was planning on keeping going until I have 20 wins, regardless of the number of losses. I CAN'T count lost games towards stall rating, because I have no way of knowing how long it would have taken me to win.
I was also figuring that I just wouldn't be able to test cards in which I lost most/many games, since the data wouldn't be accurate & it'd be frustrating.

@Jmdt: having the AI play USEM would probably be more accurate for testing USEM, but I'm hesitant to try it for a couple reasons:

-I want to be consistent with the data I've already gathered
-I'm planning on testing other decks after USEM, and the AI can't play all decks as well as it can play USEM
-I can give the AI an illegal deck. I haven't used this yet, but it might be handy later on for determining quanta requirments etc.
-It's much more satisfying to win. Sounds petty, but there's a lot of cards to get through and I'll never make it if it pops up "you lose" at the end of every game.
-When I'm playing USEM (or another deck), I can test to see how changes in play style affect ttw.

It could be very interesting to make a "skill rating" or somesuch though--see how much the stall rating changes when in the hands of a human instead of the AI. Could even be useful for making AI improvements.

[badivan1]

How can you calculate the stall rating of a card if it performs differently against some decks and not others ?

For instance, how would you calculate the stall rating of a Titanium/Diamond Shield ?

[ratcharmer]

How can you calculate the stall rating of a card if it performs differently against some decks and not others ?

For instance, how would you calculate the stall rating of a Titanium/Diamond Shield ?

Pretty much every card differs in effectiveness by opponent. Thus stall rating is dependent in part on the deck you're testing with. So far I have only been testing with USEM, but only because I haven't gotten to other decks yet (this is quite time consuming) not because I don't want to test other decks.

Eventually I hope to calculate an "average stall rating" by averaging the stall rating of a given card versus a wide variety of opponents.

[silux]

Stall cards effectiveness is hard to calculate in a deck.
I'd pick a random opponent then I'd run experiments on every card

example(not realiable datas)
maxwell demon(2 resilience)
effective
34/96 creatures
maxwell +rage potion
67/96

Squid(2 resilience)
effective
90/96
only 3 or 4 creatures

However I'd focus first against fake gods, since that is the best target for stall grinding

[ratcharmer]

Stall cards effectiveness is hard to calculate in a deck.
I'd pick a random opponent then I'd run experiments on every card

. . . that's why I'm evaluating the stall cards on their own and not in a deck?

However I'd focus first against fake gods, since that is the best target for stall grinding

Trying to run against the randomized opponents of the AI would not work well at all. One of the major concerns is that I would often be playing against decks that are stall decks in and of themselves--if the opponent is not trying to win quickly then looking at the ttw becomes largely useless.

How exactly I'm calculating all this can be seen behind the spoilers in my opening post. You can also see the results so far.

That being said, please note that I have already done testing on 23 cards (or different uses of cards). I am not going to throw all that out now because someone wants to look at it a different way.

EVERYONE PLEASE, I appreciate the enthusiasm, but if you wanted to have input on how to set up the study the time for that was three weeks ago, not now. If you want to help, then I could desperately use someone else testing cards besides just me. If you really want a test run in a specific way, then feel free to start a different study.

[Bhlewos]

I've been testing this deck (http://elementscommunity.org/forum/index.php/topic,10503.0.html) and I've gotten a 177 | 23 ratio (88.50%). I know it was ttw testing and not your stall testing but I just wanted you to know I've already had experience with this stuff. Which deck would you prefer I use, if any?