Forums Tyrant

SimTyrant - The simulator that runs from your browser! page 17

869 posts

Flag Post

Hello!
Ok i have this weird problem. I use Internet Explorer 9. I cannot understand why i’m not able to see any Speedy Achivements on the achivement list.
I screenshot it to you : http://imageshack.us/f/17/speedyui.jpg/
I hope you can help me.

 
Flag Post
Originally posted by MagoMerlino:

Hello!
Ok i have this weird problem. I use Internet Explorer 9. I cannot understand why i’m not able to see any Speedy Achivements on the achivement list.
I screenshot it to you : http://imageshack.us/f/17/speedyui.jpg/
I hope you can help me.

Yeah, this looks like an IE issue. I can replicate this problem in IE8.
I’ll look into this.

EDIT: This issue also affected Honored achievements. It should be fixed now in SimTyrant v1.102

 
Flag Post

v1.102 seems to be broken

JavaScript error:
alpha is not a function
on line 287
for http://haileon.com/SimTyrantJS
Browser CodeName: Mozilla
Browser Name: Netscape
Browser Version: 5.0 (Windows)
Cookies Enabled: true
Platform: Win32
User-agent header: Mozilla/5.0 (Windows NT 6.0; rv:12.0) Gecko/20100101 Firefox/12.0
SimTyrant version: v1.102
 
Flag Post

…And fixed in v1.103

 
Flag Post

thanks bub, you’re a dear

 
Flag Post
Originally posted by Moraku:
Originally posted by MagoMerlino:

Hello!
Ok i have this weird problem. I use Internet Explorer 9. I cannot understand why i’m not able to see any Speedy Achivements on the achivement list.
I screenshot it to you : http://imageshack.us/f/17/speedyui.jpg/
I hope you can help me.

Yeah, this looks like an IE issue. I can replicate this problem in IE8.
I’ll look into this.

EDIT: This issue also affected Honored achievements. It shuould be fixed now in SimTyrant v1.102

Fixed, thanks.

 
Flag Post

Moraku, do you have your previous versions’ code somewhere? I’d be interested to see how the code’s evolved over time w/ diffs; there are changes you make that I would like to see how you did it (and did so quickly!).

Edit: Thanks

 
Flag Post
Originally posted by purei:

Moraku, do you have your previous versions’ code somewhere? I’d be interested to see how the code’s evolved over time w/ diffs; there are changes you make that I would like to see how you did it (and did so quickly!).

I don’t keep much archive, but here’s a few that are readily available:

www.haileon.com/SimTyrantJS_v1090
www.haileon.com/SimTyrantJS_v1092.zip
www.haileon.com/SimTyrantJS_v1095.zip
www.haileon.com/SimTyrantJS_v1097.zip
www.haileon.com/SimTyrantJS_v1100.zip

 
Flag Post

still too slow loading for convenient usage :(

PageSpeed Score of 0 (out of 100) – never seen 0 actually – this is the first site :)
https://developers.google.com/speed/pagespeed/insights#url=http_3A_2F_2Fhaileon.com_2FSimTyrantJS&mobile=false

 
Flag Post
Originally posted by sss1:

still too slow loading for convenient usage :(

PageSpeed Score of 0 (out of 100) – never seen 0 actually – this is the first site :)
https://developers.google.com/speed/pagespeed/insights#url=http_3A_2F_2Fhaileon.com_2FSimTyrantJS&mobile=false

Some of that is because of ‘permanent page redirects’, which just means ‘make the url exact’. try this url: https://developers.google.com/speed/pagespeed/insights#url=http_3A_2F_2Fwww.haileon.com_2FSimTyrantJS_2F&mobile=false gives a 15/100.

 
Flag Post

Well, if it makes you any happier, here’s the PageSpeed score after GZIP.

https://developers.google.com/speed/pagespeed/insights#url=http_3A_2F_2Fwww.haileon.com_2FSimTyrantJS_3Fxczvxzcv&mobile=false

 
Flag Post
Originally posted by purei:
Originally posted by sss1:

still too slow loading for convenient usage :(

PageSpeed Score of 0 (out of 100) – never seen 0 actually – this is the first site :)
https://developers.google.com/speed/pagespeed/insights#url=http_3A_2F_2Fhaileon.com_2FSimTyrantJS&mobile=false

Some of that is because of ‘permanent page redirects’, which just means ‘make the url exact’. try this url: https://developers.google.com/speed/pagespeed/insights#url=http_3A_2F_2Fwww.haileon.com_2FSimTyrantJS_2F&mobile=false gives a 15/100.

I save the page and run locally. On my 18-month-old $300 laptop, I can get up to 600 sims per second if I am not running anything else. Otherwise, it is between 150 and 350 per second. I regularly run 100k sims on decks, but I have never tried 1mil.

But if 10 people with the same version of the sim all run 100k simulations and upload the results to the Fansite, then the Fansite can add the results together and then we have 1mil sims. Statistically, this is completely valid because each simulated battle is independent of every other battle, so the probabilities can be added together. (I KNOW that I did NOT use the proper technical terms in the previous sentence, but I can’t think of the precise and “proper” way to explain it. Nevertheless, the concept I am trying to communicate is still true. I apologize for not remembering the correct maths terms.)

In any event, Moraku’s SimTyrant is clearly reliable enough to use as our new standard. The only problem is the low number of simulations. By summing/averaging results from multiple users, that problem is solved. And it is a problem, especially for things like Raids and Quests because the enemy deck changes so often that we need millions of simulations to have accurate results. In a different thread, Moraku discussed this and posted a link to an online calculator.

</text wall>

 
Flag Post
Originally posted by hunterhogan:

I save the page and run locally. On my 18-month-old $300 laptop, I can get up to 600 sims per second if I am not running anything else. Otherwise, it is between 150 and 350 per second. I regularly run 100k sims on decks, but I have never tried 1mil.

100k sims gives trustworthy resolution ±.5% if it’s a 50% deck and ±.2% if the deck’s a 95% deck. Refining those results with millions of sims — I feel — is less important than a variety of decks that have good winrates. Not similar decks with a few cards swapped – a Zarak deck w/ a couple different cards is still a deck I can’t build. :(

Also, it’s not that sss1 objects to SimTyrant’s accuracy, it’s that he doesn’t trust the method of inputting the s because unless huge steps are taken, it’s possible to input an Orbo-spam deck and say it has a 100

And it is a problem, especially for things like Raids and Quests because the enemy deck changes so often that we need millions of simulations to have accurate results.

I understand how this feels intuitive, but this isn’t true.

I talked to an engineer friend about this to try to ferment my math until it’s digestible:

  • Say a fight against a fixed deck is a plinko board / pachinko machine – the ball falls in a ‘win spot’ or a ‘loss spot’, and there are possibly millions of spots at the bottom that are all possible paths a fight could take.
  • But! For the user of a simulator, out of this pachinko machine comes a ‘win’ or a ‘loss’, even though you know there’s a fixed number of spots the ball lands in … you don’t get to count all of them, you just run the pachinko machine 100k times and get the average win rate.
  • The feeling is that raids are really first pulling from a huge vault of pachinko machines, each with different win/losses at the bottom, so you need to sample way, way more times.
  • Another way to say it is that you feel as if you have to characterize each machine with many runs each. But Simulators are pulling a new (random!) pachinko machine every time and testing it once then putting it back.
  • The slick part is statistically, with a binary output, we don’t have to care about how big the world of outcomes is, just how many times we’ve tested
  • There are T (huge number) of total possible outcomes, including all rolls, in the whole space of a raid/fixed deck, and W of them you win. W/T is the true win rate of your deck in that system and it doesn’t matter how big T is, our sample of 100k times is as valid in a world of T=10^9999 outcomes as it is in T=10^10. Yay math!

tl;dr When you have a binary result (win or loss) there’s no such thing as a sample probability (average win against raids) that has wider/worse variance than another output of the same probability (say, average win against mission 1) unless you’ve sampled a different number of times.

 
Flag Post

Well, it’s that time again.
SimTyrant now has the new event missions:

Secret’s Return
Secret’s Reckoning
Secret’s Judgement

 
Flag Post
Originally posted by Moraku:

Well, it’s that time again.
SimTyrant now has the new event missions:

Secret’s Return
Secret’s Reckoning
Secret’s Judgement

I think they wiill change missions because “Tabitha, Lance Rider(10)” give “too good” results :)

 
Flag Post
Originally posted by purei:
Originally posted by hunterhogan:
(1. Fake simulator results)
(2. Statistics)

I was not aware of the fake results problem. A pesky problem, but it seems like there are some possible solutions.

  1. Bogus results can be replaced by someone running more simulations than the current results (at least that’s the how the old system worked)
  2. I can’t recall the name of the technology, but it’s something like incremental keys.
  3. If submited results are averaged, then outlying results can be discarded.

I thought I had some other ideas, but my creativity suddenly fell out of my brain.

Statistics
I will readily admit that I don’t know as much math as I wish I knew. If I understand the plinko board analogy, I think there is a small but important difference. On a plinko board, the assumption is that each path is identical to each other path. I think that raids are different because each deck iteration represents a plinko path with different probabilities than other paths.

Does that change the analysis?

 
Flag Post
Originally posted by hunterhogan:
(1. Fake simulator results)
(2. Statistics)

a) Sim Results – The fake submissions started out as playful, but they can be used by jerks that just want to ruin something.

1. Part of the submission is how many sims you did, so that can be faked to be billions.
2. I don’t know this one either, but since Javascript is in plaintext, anyone can discover what sekrit methods are being used to submit, and then use them as well. (Which is how this happened, EvaluateDecks was closed source.)
3. Fake submissions can be spammed, so the average can be quickly pushed in a fake deck’s favor.

sss1 could do some sort of trusted-submitter set up, but again that can be abused, and policing it is a waste of time. People seem to be submitting a link to SimTyrantJS in the comments that lets the reader load & run their own sims (you can copy/paste the ?<text> part onto your URL). Even 1000 sims is enough to get w/in ±5% (3*sqrt(p*(1-p)/n)) and convince you that their stated % is probably valid.

b) Stats – plinko paths are not all the same: the outer two need you to fall the same direction every level, but the inner ones can be reached many ways by bouncing back and forth. In my analogy one plinko board represents all deck arrangements, of both sides, and all of their outcomes. One slot per outcome, and each slot says won / lost / drew.

You are right, each outcome’s probability of showing up is not the same, but they are independent and so you can add their probabilities (the outcome where your flying activates every time, but you still lose, might be very unlikely) and they will all sum to 1.

That is for a fixed deck. So if you have a raid deck, then you have millions of plinko machines that each represent the outcomes for your deck vs. 1 raid deck. And again, the outcomes across machines are independent of each other, so you can add the probabilities all up across all machines (and they’ll sum to 1).

W/ a raid deck we’ve increased the # of possible outcomes to some ludicrous number, but, there was some p probability that when we pick an outcome from all outcomes it was a win or a loss, so the number that we test is all that matters when hunting for p, and not the # of outcomes.

edit:
c) Quests – Hunter, this is where we have more of the issue you were referring to, I believe. You have to fight the same deck a bunch, so it’s arguable that what would be best here is a deck that’s 55% against all possible quest decks, instead of 10% against half and 100% against the other half.

This one is harder to determine, and would require X decks being built, and Y tests against each, and then some sort of minimizing function across the space of the X decks. We talked a bit about this in another thread (the Latin hypercube stuff). I think a Y of 1000 would probably be alright for an estimate (it’s ±5% at worst so that’s basically 10 levels of strength with an error of ±1 level) Hmmm.

 
Flag Post
Originally posted by hunterhogan:

  1. Bogus results can be replaced by someone running more simulations than the current results (at least that’s the how the old system worked)
  2. I can’t recall the name of the technology, but it’s something like incremental keys.
  3. If submited results are averaged, then outlying results can be discarded.

There is no need to do complex calculations to determine fake results; you can simply have a log of submissions for a deck. (It may take up a lot of space over time without occasional wipes, however.)

If someone cares enough about finding a quality deck with a reliable simulation, it is likely that person would bother critically reading relevant information. People who ask questions like “what mission do I use this deck on to get the achievement?” or “how come this 1-2-3 deck works so poorly on auto?” aren’t worth catering to.

 
Flag Post

Gore Typhon related bug

An infused Wasteland Skimmer will rally itself in the game. The rally is ignored in the simulator. WS should rally itself in below example, but it’s all infused.

Turn 5 begins for Lucina the Wicked [7 HP]
Wasteland Skimmer (0) [0/5/0] reduces its timer
Wasteland Skimmer (1) [0/5/1] reduces its timer
Lucina the Wicked [7 HP] plays Wasteland Skimmer (2) [0/5/2]
Lucina the Wicked [7 HP] enfeebles Typhon’s Beast (1) [1/6/3] by 1
Turn 5 ends

So, self-faction targets must get swapped to Bloodthirsty upon infusion, or BT at least gets added. I’ve been trying to test if it can also rally imperials post-infusion, but my tests are inconclusive.

 
Flag Post
Originally posted by purei:

Gore Typhon related bug

An infused Wasteland Skimmer will rally itself in the game. The rally is ignored in the simulator. WS should rally itself in below example, but it’s all infused.

Turn 5 begins for Lucina the Wicked [7 HP]
Wasteland Skimmer (0) [0/5/0] reduces its timer
Wasteland Skimmer (1) [0/5/1] reduces its timer
Lucina the Wicked [7 HP] plays Wasteland Skimmer (2) [0/5/2]
Lucina the Wicked [7 HP] enfeebles Typhon’s Beast (1) [1/6/3] by 1
Turn 5 ends

So, self-faction targets must get swapped to Bloodthirsty upon infusion, or BT at least gets added. I’ve been trying to test if it can also rally imperials post-infusion, but my tests are inconclusive.

Ah yes, I forgot to change the faction-specific skills to target Bloodthirsty instead…
Fixed in SimTyrant v1.105

 
Flag Post

Could you add the ability to run sims using the commanders of AI tile slots? It’s impossible to get an accurate simulation without those commandeers – especially, when they have Weaken 2, Protect All 2, and Counter 2 with 22 life.

 
Flag Post
Originally posted by Anthonyk747:

Could you add the ability to run sims using the commanders of AI tile slots? It’s impossible to get an accurate simulation without those commandeers – especially, when they have Weaken 2, Protect All 2, and Counter 2 with 22 life.

You already can. The easiest way to do this is to just surge the Mission. If you like, you can also look up the Mission Commander’s ID and notate it as Typhon Eclipse[1209] for example, and then manually enter every card in his deck…

 
Flag Post

But how do you look up the commanders used in the tiles? I know how to find Missions, Raids, Quests, etc. but not Tile Commanders.

 
Flag Post
Originally posted by Anthonyk747:

But how do you look up the commanders used in the tiles? I know how to find Missions, Raids, Quests, etc. but not Tile Commanders.

Probably the best way to do this is by writing down one of the cards in the Tile Commander’s deck and then look that card up on Fansite. It’ll tell you which missions contain that card in the mission deck.

 
Flag Post

But how do you look up the commanders used in the tiles? I know how to find Missions, Raids, Quests, etc. but not Tile Commanders.

For example, if your tile’s slot 1 is Mission 228, then the opponent’s commander is Typhon Eclipse. When you mouse over the commander in Fansite, it shows “Card: 1209”.
http://tyrant.40in.net/kg/mission.php?id=316

In this case, the commander is Typhon Eclipse[1209].