Hey guys. (This will be lengthy, hahahaha)
It's nice to see that the whole idea of FGei actually survived and is still being used to establish some sort of "standard" when it comes to FalseGod-decks. You know, as I already posted in the original FGei-thread, "back in the days" (lol) people would still judge FG-decks by turns to win and highly subjective estimations such as "faster than" ... the FGei was meant to put an end to all that and I am certain that in the two studies conducted at the time, it sucessfully did. FGei is a nerdy effort to "scientifice" a minmaxing approach to the grinders everyday business. The fun side-effect is that the "race for the best deck" can actually be quantified and posted in the forums. 8)
Reading through this thread, I sometimes had to laugh pretty hard. Somewhere along these lines is what what the historians of the year 3000 will sort of sound like when they are discussing their findings of our forsaken civilization. I am glad some of you have bothered to read my (sometimes a bit wild) documentation. I didn't want to leave something that worked "somehow .. who knows how" even though I may still have. But I totally understand what you mean though: Devising the Statmasta had taken me soooo many hours of tedious bug-hunting and perfecting every single part of the spreadsheet ... at the end it would take me like 2 hours to get back "into" it when I wanted to change a minor detail. (That having said, I am no IT-guy or mathwiz ... these people will certainly boast a broad laugh at this point. For me, learning Excel and Macroes and all that bit by bit while devising the spreadsheet was quite a "job".)
Now I don't want to sound like an old fogey or anything. I am just posting in order to hopefully help you out if you ever decide to do a study like this again (and if the FG-battle mode as it is known now survives in 1.4 of course). So here are my thought on FGei1.32 revamped and some select post from this thread:
Last time I checked, in 1.32 only Akebono and Osiris had gotten a new deck: 2 shards were added or something. Other than that, everything is still in place, right?
<- This is the weak spot for all that follows from here. If I overread any major changes then this could change a whole lot. I apologize for going on an on if I were simply to lazy to read that [insert major change] in the patch notes.So as long as there is ...
- no brandnew god
- no complete revamp of FalseGod-decks
- no grave change to the card-spin-wheel
- no grave recalculation of the resell-value of cards
- no change in win or EM-payout or loss electrum (all rather minor but notable in fact)
... I can assure you: The Statmasta is still solid.
If you had been there when I cross-checked results like a hundred times most of the time even using a good old CALCULATOR as one instance of quadruple-checking the values, you would know what kind of "solid" I am talking about here.
The Statmasta adresses every single bit and piece of electrum + electrum per hour there is in the game adequately (read: 99.5% perfectly) with the exception of:
1. Exact resell value of cards-> some cards are marginally more expensive than others
-> gods with a high number of rare spins (prominent: Divine Glory) should have a notably higher resell value
=> The first issue is absolutely neglectable.
The second issue has always bothered me quite a bit. One would need the theoretical spinrates for every single card of a god to recalculate the actual value. Assuming that with Divine Glory (THE rare god) every second spin is a rare (which is quite generously assumed I think), the resellvalue in the Statmasta for DivineGlory should not be 1160 but rather 1230 ... a 6% increase of the most significant source of electrum gain in FG-farming (factoring in all the other sources of income probably coming out somewhere 4%-5% electrum gain not accounted for) with this single, most outstanding rare-farm-god. "Notably higher"? Yes. Crucial? Not if you consider the fact that you will only meet him 5.17 times in a set of 150 games. If you now assume 3 gods of this high-rare-profile type (DG, Octane, Osiris) you wind up somewhere around a 0.5% higher FGei than actually shown:
Instead of 9000 FGei(cn) you should have around 9045 ... deliberately bragging here: That would be the single most prominent flaw that bothered me.
2. EMs with more than 100HP
-> this issue is addressed in the guide-thread and the manual recalculation workaround (option a) of the missing electrum is fully accurate as long as the finishing HPs for these EMs are recorded seperately and accurately. A bit annoying to keep track of it but not impossible and, more importantly, only necessary if you do play a deck relying on high HPs.
Having tried to emphasize the degree of solidness of the Statmasta, I can understand that a convenient google-sheet would be so much better to use in a joint-effort study. In the first big study we did with the Statmasta, I manually compiled all the files of all the testers into one big file for a single deck to generate a large sample. It was quite a pain in the ass indeed. But it was solid. Unfortunately, simply exporting the formulas to google-docs will most likely result in more or less crucial errors. At the time I even had to struggle with compatibility issues between different versions of Excel (not even talking about OpenOffice, LibreDocs and all that here) ... eversince I have the highest respect for IT-developers: Things should be a lot easier but for some stupid reason they never are. I am not saying that it can't be done or that the current version of the docs-sheet is flawed (because I didnt look at it). It sure would be awesome to have a working version of it online.
1) STATMASTA is inaccurate, if not useless. Its core calculations rely on figures that are both wrong and outdated.
2) The Community has no information that is both accurate and current on individual FG drop rates for Akebono or Osiris.
Good thing you crossed out the first statement. As stated above, everything seems still in place.
Akebono and Osiris ... yes ... the few cards that changed are of course also changing the drop-rates. But the change is going to be somewhat marginal. If you followed my calculation on how DivineGlory payout is in fact neglectably incorrect (incorrect nevertheless), then you will guesstimate how significant these few cards will be for a final "sum it all up" FGei of a deck. A correction of the droprate would still be very nice of course.
I thought FGei was simply a method to test and calculate efficiency. I'm surprised using a method that requires normalization being used since the number of tests that are required to reduce the percent error are significantly higher than a fixed count. I figured a hard cut off would be used or something like that. (again should have read more.)
Absolutely right Keeps. Sample sizes in both indivual player test as well as joint-efforts have at all times been too low to actually do normalization. The highest sample I ever saw was like 600 games ... and that is still too low if you ask a mathematician.
Now the Statmasta "unlocks" normalized stats as soon as you have played 3 games against each god. Ridiculous!
The problem has always been the harsh reality of FG-grinding ... for me personally, 180 games with a single deck was as high as I ever got before I totally got sick of it and told myself that I was supposed to have fun instead of being anal about shit like this. What comes into play here is "common sense perfectionism": Can't do more than like 200 games before you puke? Ok then ... let's at least level out the wrinkles on that level. In calculating the FGei(cn) for you, the Statmasta "normalizes" on many levels: It will assume you played each god equally often. It will assume normalized cardspin-rates in the first place (cf DarkWeavers stats). It will even "normalize" the time played total based on every individual record for each FG (as expressed in the FGei(c)) before recalculating the FG(cn) ... of course, if a puny little 3 games are what these calculations are based on for a single god, then the "accuracy" for this gods portion of the pie is at best 33.3% (the weight of each game). If you consider that 1 of these three games might as well have gone the other way, then you know there is a problem.
Lessons to take from this:
1. Testers should know their deck by hard! Play at least (!) 100 games with a deck before actually testing it. You have to know every single trick in the book and play it right in order to reduce that "misnormalization"-risk to pure luck of the draw. This inflicts heavily on the time needed for a game as well. Having to reconsider what to do is not acceptable if you really want a valid FGei. The FGei is a grinding-index ... be a grinder already!
2. Well known: Create samples that include at least 10 games against each god. This refines the "accuracy" down to at least 10%. This is what joint efforts are for. There is not really a point in testing 20 different decks with like 7 people. All of them being grinders, they should know the eligible decks already ... now they want to know for sure. Preliminary decktesting with like 80game samples is for the deck-sections. A proper study gets in like 300+ games done by at least 2 dedicated pros with that deck.
(You could of course always fall back to "just a personal index" but then you are stuck with "well ... but GamerDude doesnt know how to play this, I would have done better yadadada" again. Which is what the FGei as an index for studies meant to eradicate. Getting two top-guys on the job yields: "Personal but as frigging damn good as it gets with this deck!")
Spin chance (both average and individual) is known. The GDoc spreadsheet (as of last time I heard anything about it) was using average spin chance rather than individual spin chance. It's not a big deal overall [...]
In real world situations, it makes a tiny difference in total money but doesn't really tend to change the rankings much if at all. It would still be better to use the individual chances rather than the average chance, though, since IMO pretty much anything that improves accuracy is worth doing. Previous FGEI-calculating spreadsheets have used individual chances.
As a matter of fact it is a HUGE DEAL. The very first two versions of the Statmasta had an overall average spin-chance. This was first 33% (very long time ago) and then 40% (when the new FGs were introduced), both figures once stated by Zanz to be the average card-win-rate. The realtec-edition of the Statmasta features actual card-spin-rates and as petty as this may sound, it was THE advancement towards actual, real-life electrum gain you could expect from a deck. You can almost forget all the EM-, win- and loss electrum ... FG-grinding generates electrum primarily by winning upped cards. If your spin-rate is 70% or 30% makes a huge difference here. This goes especially for the new generation of specialty decks: Specialty decks tend to cater to specific gods ... You would be surprised what happens when you introduce the actual numbers.
DarkWeavers numbers were solid and matched empirical data as generated in two seperate studies. From what I understood back then, it's even not all that hard if you know math. Find somebody who does and he will give you the numbers quite quickly. But as long as the FG-decks haven't changed, these numbers are just fine. There even is some tool somewhere: You just enter the deck and it yields spin-rates for each card.
So if anyone has any good ideas on how to convince zanz (or someone else with access, if there is anyone else) to post the current code for the spin algorithm, that would be fantastic.
Excellent ideas. From what I gather, Zanz has records on any- and everything. I believe he should even have the raw numbers for each god ... no algorithm needed here. If he could just hand over the exact numbers as empirically generated from thousands of games.
It only makes sense... you added a small amount of time to your play, cost yourself a little money, you don't just magically get more electrum / hr.
EDIT: Scrap that. Sorry guys. Not having been in the matter for a long time, I fell for common sense logic here. There are in fact (sort of extreme) instances where this is possibleYes. You lose a game, be it by skipping or losing it the hard way, your FGei drops. Simple as that. Of course, if you have already played like 400 games it will not drop significantly but it will drop. A skip is usually accounted for as a ~1sec game. Of course this will hardly inflict your FGei because losing 30 electrum in 1sec is nothing when weighed against winning 1160 electrum in 120 secs. That's why people skip.
And accouting for the loss adequately by weighing the time played against electrumg gained is what the Statmasta (when used correctly) does. Always.
The mere fact that the google-sheet used here sometimes shows an increase in FGei after entering a loss/skip tells me that it is (was?) definitely not working correctly.Fun experiment: Take the raw data from this spreadsheet, toss it in the generally-accepted-to-be-functioning-correctly StatMasta spreadsheet, compare the results, go from there.
Also, I still think someone (who isn't me) should just pull the raw data from the GDoc and plug it into statmasta and compare.
Yes.
When v1.4 arrives, FGei will be a thing of the past. Over. Done. Useless. Kaput.
Although I can somehow anticipate and welcome the "adventurous unpredictablility" of the intended 1.4 system, it would be a shame if the great tradition of FG-farming simply vanished in a blur of randomness. Crafting and refining decks to that specific purpose has always been the "Kings discipline" of Elements and a 1.4 like that would be like introducting a new idea to the 100m mens: "Every now and then somebody randomly shoots a few contestants in the leg"
Then again, who said you can't craft and refine decks to the specific purpose of first, beating a minion and second, beating the actual false god. FGei translates as "electrum gain per hour on average" ... this idea can never be a thing of the past unless electrum gets taken out of the game. The Statmasta as existing right now however ... certainly gone if 1.4 decides to randomly shoot people in the leg.
