*Author

moonlighting

  • Guest
Re: FG Efficiency Study - Applying statistics to all of the myths https://elementscommunity.org/forum/index.php?topic=38236.msg397712#msg397712
« Reply #336 on: September 23, 2011, 06:18:38 am »
did anyone notice, that the probability a certain god occurs are not equal?

from all those games listed in the OP, e.g. akebono, ethernal phoenix etc. are always met fewer than other gods.

anyone can confirm that all gods should appear at a same frequency?

Offline Jangoo

  • Sr. Member
  • ****
  • Posts: 877
  • Reputation Power: 0
  • Jangoo hides under a Cloak.
  • New to You
Re: FG Efficiency Study - Applying statistics to all of the myths https://elementscommunity.org/forum/index.php?topic=38236.msg397873#msg397873
« Reply #337 on: September 23, 2011, 01:18:21 pm »
did anyone notice, that the probability a certain god occurs are not equal?

from all those games listed in the OP, e.g. akebono, ethernal phoenix etc. are always met fewer than other gods.

anyone can confirm that all gods should appear at a same frequency?
This is from the "all games, all players, all gods"- file I made from our 6755 games:

FGnumber of games played
(6755 total)
Akebono
Chaos Lord
Dark Matter
Decay
Destiny
Divine Glory
Dream Catcher
Elidnis
Eternal Phoenix
Ferox
Fire Queen
Gemini
Graviton
Hecate
Hermes
Incarnate
Jezebel
Lionheart
Miracle
Morte
Neptune
Obliterator
Octane
Osiris
Paradox
Rainbow
Scorpio
Seism
Serket
223
269
236
239
251
224
222
225
218
223
226
240
260
235
248
230
244
229
236
222
199
234
211
215
240
240
277
236
203
I think it's safe to say that any perceived discrepancies concering the encounter-rate are purely subjective.  8)


kirchj33

  • Guest
Re: FG Efficiency Study - Applying statistics to all of the myths https://elementscommunity.org/forum/index.php?topic=38236.msg397942#msg397942
« Reply #338 on: September 23, 2011, 04:44:27 pm »
Just a few notes,

The stats have now all been updated through the STATMASTA Realtec engine which produces more accurate stats based upon real card-spin rates (except for Voodoo Panic - HG1&O mod).  These have been updated as Jangoo recently delivered them to me.  We could really use 500 total games worth of Fast-Draw Ghostal data since it is the current "champion".  That, along with any untested decks, should be highest priority right now for testers.

We are all looking forward to the skip data which will come shortly!

Edit: All decks now added with Realtec stats and quantified in order of performance.

Offline Jangoo

  • Sr. Member
  • ****
  • Posts: 877
  • Reputation Power: 0
  • Jangoo hides under a Cloak.
  • New to You
Re: FG Efficiency Study - Applying statistics to all of the myths https://elementscommunity.org/forum/index.php?topic=38236.msg398012#msg398012
« Reply #339 on: September 23, 2011, 07:16:47 pm »


We could really use 500 total games worth of Fast-Draw Ghostal data since it is the current "champion".  That, along with any untested decks, should be highest priority right now for testers.

We are all looking forward to the skip data which will come shortly!

Edit: All decks now added with Realtec stats and quantified in order of performance.
Awesome saucesome.

Concerning skip-stats, I did have a procedure in mind but I would like to share and ask what
you guys think would be best to do since I am not all that sure myself which way would be fair and safe to go:


BASICS  (this is how it will generally go down)
  • Skip stats are created by sorting out entire gods, not single games
  • The criterion whether a god is worthwhile to play is its individual FGei(c)
  • This is measured against the FGei(cn) of the deck -> Its average performance against all gods
  • If its "below", the god becomes a skip-god, if its equal or higher the god stays in
  • If designated as a skip-god, all of a gods games are set to "skip" ... yes, the few wins as well
  • All of the gods times for each game are set to 5 secs
  • The result will be a significantly increased FGei(c) and FGei(cn)

Two questions

The two spots marked in red are where I am wondering.
The second one is rather easy to answer, I think:

Is 5 seconds realistic when assuming a skip or should it be more/less?

The first red spot is part of a bigger problem:


THE PROBLEM

Now the problem with this procedure is our small sample-size for some decks.
While I think it is safe to go ahead with the 6 top-decks (except Ghostal) since all of them
have 500+ games under the belt, the stats of decks with 300 games or even less will most
likely be distorted quite a bit.
The reason is that even 300 games imply "only" 10.5 games per god on average and since
those 300 games didn't even spread out equally across the gods, some gods have to make
due with e.g. only 6 games ... not exactly a number you'd make a definite call with I think.

Now consider again the obvious that implementing skips means that the FGeis assume
your gametime according to the profile the deck has against the gods or in other words:

If you take out a god, then the FGeis assume that the time you used to spend playing
the skip god during an hour is spent playing against more worthwhile gods.


Examples for two ways this could go wrong:


1. Overrating a deck that only has <500 games

Several gods have a very high FGei(c) based on very few games.
If somebody had played some more games, it just wouldn't be this high
since those few games were a bit lucky.

-> After implementing the skips, the decks FGei(c) and FGei(cn) skyrockets!
-> It is assumed that you spend most of your time playing against gods
worth an FGei(c) of 10.000 something, when really this number is based on
just a couple lucky games.


2. Underrating a deck that only has <500 games

Several gods that should be the strength of the deck, or at least doable,
weren't played very often. Also, those few games were a bit unlucky.

-> After implementing the skips, the FGei increase simply isn't very great.
(It won't ever plummet you know, so not awarding what "should be"
is the problem here.)
-> It is assumed that you spend most of your time playing against gods that
are actually quite mediocre opponents for this deck. Some of the true
cash-cows got sorted out or just dont pull their weight based on a few
unlucky games.


SOLUTIONS

When it says "below" (in red, up there under "basic procedure") I don't have
a very strict "below" in mind as in "just one FGei-point below and you are out".

Traditionally, only the gods that are just absolute fail get sorted out.
Since we have very nice, accurate data here, we can certainly be a bit more picky
so I was thinking something along the lines of:

If a gods individual FGei(c) is more than 20% below the FGei(cn), the god becomes a skip-god.

Now even a buffer-percentage like this doesn't exactly solve the above problem but it can
certainly help alleviate it a bit.


The questions part 2

  • This 20%-number is of course completely arbitrary, so I am wondering which number would be a good choice.
  • Should decks below a certain number of games even be skip-implemented?
  • Would a staggered skip-implementation be a good idea, where decks >500 games = 20%,
    deck >300 games = 30%, decks <300 games = 40%  ?
  • Any other ideas concering the above problem?
Concerning the staggered implementation:

While it helps alleviate parts of the problem, it also means that any decks with a smaller sample-size
will not profit from the skip-implementation as much: The bigger the margin of tolerance the less pointed
the assumed performance-increase will be ...


kirchj33

  • Guest
Re: FG Efficiency Study - Applying statistics to all of the myths https://elementscommunity.org/forum/index.php?topic=38236.msg398022#msg398022
« Reply #340 on: September 23, 2011, 08:00:37 pm »
...stuff
5 secs/skip is perfectly reasonable.

I would not even consider putting any of the decks with less than 500 games through your machine.  As this threshold is reached, more decks can be added.

20 or 25% seems reasonable.  I would play with it a little if I was you and see how it turns out.  I think a percentage-based number is the best way to compare decks uniformly as this creates a nice broad smoothing effect.  Once the standard is established, decks such as Fast-Draw Ghostal can easily be entered in to quickly output results and add to the study.

Offline Onizuka

  • Legendary Member
  • ******
  • Posts: 2292
  • Country: us
  • Reputation Power: 0
  • Onizuka hides under a Cloak.
  • Donuts!
  • Awards: Slice of Elements 6th Birthday CakeSlice of Elements 5th Birthday Cake1st Person to Escape the TempleSlice of Elements 4th Birthday CakeChampionship League 2/2012 2nd PlaceWeekly Tournament WinnerSlice of Elements 3rd Birthday CakeWinner of the Harry Potter PvP House Cup - HoH5th Trials - Master of EntropyWeekly Tournament WinnerWeekly Tournament WinnerWeekly Tournament WinnerSlice of Elements 2nd Birthday Cake
Re: FG Efficiency Study - Applying statistics to all of the myths https://elementscommunity.org/forum/index.php?topic=38236.msg398042#msg398042
« Reply #341 on: September 23, 2011, 08:35:33 pm »
Most of my skips are 3-4 seconds. But 5sec/per is also fine.
You're just as selfish as I am. You're just not as good at it yet.

Offline 10 men

  • Sr. Member
  • ****
  • Posts: 987
  • Country: de
  • Reputation Power: 0
  • 10 men hides under a Cloak.
  • Honesty is the privilege of the infallible
  • Awards: 6th Trials - Master of TimeWinner of Draft #3 - PvP EventWeekly Tournament WinnerWeekly Tournament Winner5th Trials - Master of Time4th Trials - Master of TimeWeekly Tournament WinnerWeekly Tournament WinnerShard Revolution WinnerSlice of Elements 2nd Birthday CakeWar #2 Winner - Team Entropy
Re: FG Efficiency Study - Applying statistics to all of the myths https://elementscommunity.org/forum/index.php?topic=38236.msg398150#msg398150
« Reply #342 on: September 23, 2011, 11:22:37 pm »
Ok, first of all I have reached 500 games with Ghostal (those stats include the 206 games I've played previously):

  deck     Fast-Draw Ghostal   
  players     10 men   
  version    1.285 
  win-rate     69,4%   
  $ track ?    enabled 
  win-rate (n)     69,26%   
  games    500 
  Score/h     938   
  win-loss-(EM)     347-153-(259)   
  Score/h (n)    949   
  time (h:m:s)     35:46:43   
  FGei[c]* (http://elementscommunity.org/forum/index.php/topic,14626.0.html  6167   
  min/game     04:18   
  FGei[cn]* (http://elementscommunity.org/forum/index.php/topic,14626.0.html  6276   
      Statmasta™realtec                                                               

  score/h  FGei(c)   Statmasta™realtec    wins losses skips EM/Wins
   1351
   852
   655
   -335
   1029
   1358
   -219
   1528
   1037
   1374
   702
   1089
   421
   1076   
   810
   1278
   155
   1187
   1579
   938
   1388
   -411
   735
   877
   1451
   358
   620
   801
   1509
   10767
   4706
   5128
   2024
   5585
   11032
   2474
   7108
   7829
   7412
   5024
   6948
   4482
   8629   
   4900
   7248
   2454
   6622
   7340
   5315
   7701
   3465
   7461
   6954
   8417
   2528
   4626
   4815
   9971
Akebono
Chaos Lord
Dark Matter
Decay
Destiny
Divine Glory
Dream Catcher
Elidnis
Eternal Phoenix
Ferox
Fire Queen
Gemini
Graviton
Hecate
Hermes
Incarnate
Jezebel
Lionheart
Miracle
Morte
Neptune
Obliterator
Octane
Osiris
Paradox
Rainbow
Scorpio
Seism
Serket
   8
   13
   7
   2
   22 
   10
   3
   16
   17
   11
   10
   20
   8
   15
   11
   18
   5
   24
   16
   11
   9
   3
   10
   13
   21
   6
   16
   14
   8
    3
    7
    7
    13
    2 
    2
    14
   
    7
    1
    8
    5
    10
    3
    6
   
    11
   
   
    1
   
    14
    8
    3
    1
    9
    10
    6
    2
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
     5
     11
     3
     1
     22
     9
     2
     15
     13
     11
     10
     17
     3
     4
     7
     18
     4
     24
     16
     3
     9
     
     5
     6
     20
     5
     1
     10
     5
http://www.mediafire.com/?rzfk54cv0l6k55w

Stats from someone else would still be nice of course.

Then I also have some questions/suggestions on the Statmasta:
- Knowing how much time it takes to quit or to spin is not only interesting for skips, you also spend some time in any game on leaving, entering a new game and doing your spin. Does the Statmasta currently factor those times in? Otherwise FGeis are a little too high of course as you also always spend some time outside the game.
- Does it give you the extra 5 coins you get in the spin when the first two cards match? Ask Dark Weaver for the probabilities, I think they're not difficult if you already have the probability for 3 matching cards.
- Does it factor in the different prices of winnable cards? For example Divine Glory's spins are more valuable because she drops Miracle and Morning Glory which both sell for about 150 :electrum more than normal cards.

Quote
This 20%-number is of course completely arbitray, so I am wondering which number would be a good choice.
I did a little experiment with my data and replaced the games of one god after another with 5 second long skips starting with the least profitable and looked at which point the FGei did not increase. It increased all the way up to Lionheart (!), a god against whom I had an above average FGei and 100% winrate. So it seems that if you're really serious about optimizing money gain (and don't care about your win/loss ratio at all  :)) ), the best way is in fact cherrypicking the gods with the absolute highest profitability and playing only those. The entry cost of 30 electrum is just too low compared to the possible winnings. Here's how my stats would look with skips:

  deck     Fast-Draw Ghostal   
  players     10 men   
  version    1.285 
  win-rate     38,4%   
  $ track ?    enabled 
  win-rate (n)     41,18%   
  games    500 
  Score/h     724   
  win-loss-(EM)     192-308-(153)   
  Score/h (n)    800   
  time (h:m:s)     17:00:00   
  FGei[c]* (http://elementscommunity.org/forum/index.php/topic,14626.0.html  7236   
  min/game     02:02   
  FGei[cn]* (http://elementscommunity.org/forum/index.php/topic,14626.0.html  7452   
      Statmasta™realtec                                                               

  score/h  FGei(c)   Statmasta™realtec    wins losses skips EM/Wins
   1351
   -21600
   -21600
   -21600
   -21600
   1358
   -21600
   1528
   1037
   1374
   -21600
   1089
   -21600
   1076   
   -21600
   1278
   -21600
   -21600
   1579
   -21600
   1388
   -21600
   735
   877
   1451
   -21600
   -21600
   -21600
   1509
   10767
   -21600
   -21600
   -21600
   -21600
   11032
   -21600
   7108
   7829
   7412
   -21600
   6948
   -21600
   8629   
   -21600
   7248
   -21600
   -21600
   7340
   -21600
   7701
   -21600
   7461
   6954
   8417
   -21600
   -21600
   -21600
   9971
Akebono
Chaos Lord
Dark Matter
Decay
Destiny
Divine Glory
Dream Catcher
Elidnis
Eternal Phoenix
Ferox
Fire Queen
Gemini
Graviton
Hecate
Hermes
Incarnate
Jezebel
Lionheart
Miracle
Morte
Neptune
Obliterator
Octane
Osiris
Paradox
Rainbow
Scorpio
Seism
Serket
   8
   
   
   
     
   10
   
   16
   17
   11
   
   20
   
   15
   
   18
   
   
   16
   
   9
   
   10
   13
   21
   
   
   
   8
    3
    20
    14
    15
    24 
    2
    17
   
    7
    1
    18
    5
    18
    3
    17
   
    16
    24
   
    12
   
    17
    8
    3
    1
    15
    26
    20
    2
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
     5
     
     
     0
     
     9
     
     15
     13
     11
     
     17
     
     4
     
     18
     
     
     16
     
     9
     
     5
     6
     20
     
     
     
     5

What's important here though is to have a very good measurement of your FGei in its meaning as Electrum/hour. I would advise to implement the above mentioned features first and take only decks wit at least 500 plays.
"My big fear is that one day I may become so vain that I will quote myself in my own signature."  ---  10 men

Offline gumbeh

  • Full Member
  • ***
  • Posts: 434
  • Reputation Power: 7
  • gumbeh is a Spark waiting for a buff.
  • Nude to Elements
Re: FG Efficiency Study - Applying statistics to all of the myths https://elementscommunity.org/forum/index.php?topic=38236.msg398175#msg398175
« Reply #343 on: September 24, 2011, 12:10:54 am »
I'm not sure a percentage makes the moist sense.

Some decks do fine against a large number of gods, and only have significant trouble with, say, 6 gods. Skipping more than that is wasteful, because it's inflating the FGei by focusing on more profitable gods while needlessly costing you score and electrum by suggesting that you skip gods who are actually worth your time, even if not strictly as worthwhile as other gods you can choose.

Some decks would want to skip a larger percentage of gods - maybe there are 12 gods against which a given deck's "combo" just doesn't work. Or those gods have PC and the deck fails vs any PC. Cutting out only the bottom 20% makes the deck look worse than is realistic, because it's not skipping gods that any player of the deck knows SHOULD be skipped, even if those gods aren't as low as the bottom 20%.

Maybe there's a way this can be calculated in a machine-like manner, maybe not. For upped decks we could have a flat threshold - 2800 FGei or bust, for example. I think a more tailored approach to each deck is going to be the most correct, though. Each deck would have to be looked at individually. The question, I feel, is less about "which gods are in the bottom 20%" and more about "below which FG do the FGei numbers drop below acceptable?" Each god's FGei should be compared not only to other data points for that deck, but also to some sort of a global minimum. Most of the upped decks have 4000 FGei or more. Aiming for that, I look at this:

(https://docs.google.com/spreadsheet/oimg?key=0AlCkkeUwgmZ3dFJWVWxKQWVMcEp3eFF0aDA0RVdaZGc&oid=1&zx=u1kzjw9ytjp)
If I'm aiming for roughly "4k or bette", I can see that Decay is pretty close, but there's a big drop below Decay. So I choose Decay as my cutoff point, and everything below it is a skip. If I were 10 men, and money/score is no object, I might choose a higher cutoff point, like the gaps between Gemini and Dream Catcher, or Incarnate and Lionheart.

(the image doesn't have each bar labeled on the graph, but you can hover over them in the original google doc)
https://docs.google.com/spreadsheet/ccc?key=0AlCkkeUwgmZ3dFJWVWxKQWVMcEp3eFF0aDA0RVdaZGc#gid=1


twixy10

  • Guest
Re: FG Efficiency Study - Applying statistics to all of the myths https://elementscommunity.org/forum/index.php?topic=38236.msg398218#msg398218
« Reply #344 on: September 24, 2011, 01:21:37 am »
Maybe a system that starts with 25% below, but looks at the amount of games played against that god (the accuracy of the stats), and lowers the percentage when the amount of games are higher (against that specific god).

example: 5-10 games played has the border of 25%, 10-15 makes 20%, 15-20 makes 15%, 20-25 makes 10%

Offline gumbeh

  • Full Member
  • ***
  • Posts: 434
  • Reputation Power: 7
  • gumbeh is a Spark waiting for a buff.
  • Nude to Elements
Re: FG Efficiency Study - Applying statistics to all of the myths https://elementscommunity.org/forum/index.php?topic=38236.msg398259#msg398259
« Reply #345 on: September 24, 2011, 02:39:32 am »

Offline Jangoo

  • Sr. Member
  • ****
  • Posts: 877
  • Reputation Power: 0
  • Jangoo hides under a Cloak.
  • New to You
Re: FG Efficiency Study - Applying statistics to all of the myths https://elementscommunity.org/forum/index.php?topic=38236.msg398581#msg398581
« Reply #346 on: September 24, 2011, 05:42:47 pm »
Ok, first of all I have reached 500 games with Ghostal (those stats include the 206 games I've played previously):

  deck     Fast-Draw Ghostal   
  players     10 men   
  version    1.285 
  win-rate     69,4%   
  $ track ?    enabled 
  win-rate (n)     69,26%   
  games    500 
  Score/h     938   
  win-loss-(EM)     347-153-(259)   
  Score/h (n)    949   
  time (h:m:s)     35:46:43   
  FGei[c]* (http://elementscommunity.org/forum/index.php/topic,14626.0.html  6167   
  min/game     04:18   
  FGei[cn]* (http://elementscommunity.org/forum/index.php/topic,14626.0.html  6276   
      Statmasta™realtec                                                               
Stats from someone else would still be nice of course.
Awesome work.

Stats from at least two different players would indeed be great ... not only for this deck.
So far, it seems that in the very most cases no two players played the same deck alike ... FGeis were always shifted, sometimes significantly.


Quote
Then I also have some questions/suggestions on the Statmasta:
- Knowing how much time it takes to quit or to spin is not only interesting for skips, you also spend some time in any game on leaving, entering a new game and doing your spin. Does the Statmasta currently factor those times in? Otherwise FGeis are a little too high of course as you also always spend some time outside the game.
That is actually a very good point.

The Statmasta as such doesn't factor these times in since recording times is, in the product idea and implementation of the Statmasta,
an optional choice when taking deck-stats (hell, outside of this thread most people stick to the win-rate thing as it seems).
As such, it is merely a matter of which times the player enters on the input-sheet and there is no additional flat-rate-time coded.

In this study, testers have only recorded the times as they show up on the details screen after the game.
I am guessing the timer starts when the coin-toss is decided and ends when a player reaches 0 HPs.
So effectively the time to click "false god", to wait for the coin-toss, to maybe click "details", to click "spin",
to wait for the spin and to finally click "ok" is not accounted for.

Looking at it that way I'm thinking: "Right ... why should skips have a timer on this when wins and losses don't?"
On the other hand, in skips, you don't get anything for that time ... just lost time until you finally play a real match.
Nobody has ever skipped as aggressively as we will be simulating it here and traditionally decks have been evaluated
by their performance against all gods.
So strictly spoken, timing skips with anything higher than 0 sec would be a penalty. Perhaps a fair penalty but a penalty.

Personally, I am undecided on this one. On a sidenote, the Statmasta needs a min. time of 1 second because the
way I did the formulas it can't handle a 0 right now.

Quote
- Does it give you the extra 5 coins you get in the spin when the first two cards match? Ask Dark Weaver for the probabilities, I think they're not difficult if you already have the probability for 3 matching cards.
- Does it factor in the different prices of winnable cards? For example Divine Glory's spins are more valuable because she drops Miracle and Morning Glory which both sell for about 150 :electrum more than normal cards.
I guess these things have been considered neglectable so far.

As of now, the Statmasta assumes a flat-rate 1160 electrum for each card won.
This is the average card-sell value + 5 electrum for the triple-spin, so this coin is factored in as long as you get the card.

Card-values:
I somehow calculated the average ages ago. It certainly doesn't factor in the distribution of the actual cards won based
on probabilities. For all I know, the first usable data on that was produced by DarkWeaver and the Drop-rate study just recently.
But this could make a notable difference, even outside of the obvious cases of DG, Osiris and Octane.

Double-spins:
If we take a 50% win-rate, 2min/game deck (very good deck)
and assume the average cardwin-rate of 47.7% (DarkWeaver) we got in an hour:

30 games
15 wins
15 x .477 = 7 wins accounted for (triple-spins)
8 wins unaccounted for (double-spins)

The question is how many wins with double-spins there are in these 8 games on average?
Four? Then it would a missing ~20 FGei per hour and less for worse decks.
This is not very much but still worth putting in.


Quote
I did a little experiment with my data and replaced the games of one god after another with 5 second long skips starting with the least profitable and looked at which point the FGei did not increase. It increased all the way up to Lionheart (!), a god against whom I had an above average FGei and 100% winrate. So it seems that if you're really serious about optimizing money gain (and don't care about your win/loss ratio at all  :)) ), the best way is in fact cherrypicking the gods with the absolute highest profitability and playing only those. The entry cost of 30 electrum is just too low compared to the possible winnings.
The funny thing is: Lionheart is already well over the average and still gets to go.
How would you call this procedure? "Progressive"?

But yeah, that is pretty much the question here:
How far should this go? What should be the measurement for a skip/should play god?

The skips-stats were intended as a practical farming-guide and less as a mathematical showcase
how far you can optimize by shaving off the last bit of common sense and exploiting every god
that only has some five games played against.

Another question is: What is usability-wise doable?
Perhaps, designating as many gods as you can looks good on paper but just sucks when
in praxis you wind up having to skip 15 gods ...
Looking at e.g. Gumbehs views on that, he has a much more "down to earth" stance when it
comes to what is worth your time and what is not.

All that having said, the question how acceptable it is to sacrifice score at the gain of just
a bit more electrum isn't even being considered in all this yet.


Quote
What's important here though is to have a very good measurement of your FGei in its meaning as Electrum/hour. I would advise to implement the above mentioned features first and take only decks wit at least 500 plays.
Concerning

- Card-values, I think I have to wait for some more conclusive data on this
(and somebody who adds it all up I guess ...)
- Double-spins, I will put those in with the next update (and when I know how many of these there actually are of course)

Other than that, I don't really understand what you are getting at by "good measurement of you FGei and its meaning" ...?

As for 500 games min. to implement skips, kirch also considers this best-practice, but:
What are we going to do with the decks that are below that, and probably always will be
because nobody is going to test them anymore? Certain skips certainly will be very visible
even with few games, which is, what I thought, the staggered skip-implementation could
still be useful for.


I would not even consider putting any of the decks with less than 500 games through your machine.  As this threshold is reached, more decks can be added.

20 or 25% seems reasonable.  I would play with it a little if I was you and see how it turns out.  I think a percentage-based number is the best way to compare decks uniformly as this creates a nice broad smoothing effect. 
For upped decks we could have a flat threshold - 2800 FGei or bust, for example. I think a more tailored approach to each deck is going to be the most correct, though.

[...]

Maybe a system that starts with 25% below, but looks at the amount of games played against that god (the accuracy of the stats), and lowers the percentage when the amount of games are higher (against that specific god).

example: 5-10 games played has the border of 25%, 10-15 makes 20%, 15-20 makes 15%, 20-25 makes 10%
Yeah, I also think it should be tailored in as the skips considering each deck on its own and the best way to
optimize its performance. Otherwise, bad decks would kind of just land in the waste-basket when really
even with them you can farm (because its fun or whyever) and perform just a bit better by making the right skip-choices.

Then again, the smoothing effect kirch mentioned and that all decks are held by the same standard is  pretty important.

I find Gumbehs idea to take the accuracy/reliability of the stats for each god as a measurement is pretty awesome here:

The more reliable a gods stats are, the harsher the skip-god threshold can be.
For barely played decks, this could mean that at least for a handful of gods, there
can be made a somewhat reliable prediction whether they are worth it or not, even
if the rest of the gods better be left alone.



moonlighting

  • Guest
Re: FG Efficiency Study - Applying statistics to all of the myths https://elementscommunity.org/forum/index.php?topic=38236.msg399488#msg399488
« Reply #347 on: September 26, 2011, 02:37:28 am »
why do you skip morte and lionheart? the win rate against them is pretty high. this will be interesting maths behind...

Ok, first of all I have reached 500 games with Ghostal (those stats include the 206 games I've played previously):


I did a little experiment with my data and replaced the games of one god after another with 5 second long skips starting with the least profitable and looked at which point the FGei did not increase. It increased all the way up to Lionheart (!), a god against whom I had an above average FGei and 100% winrate. So it seems that if you're really serious about optimizing money gain (and don't care about your win/loss ratio at all  :)) ), the best way is in fact cherrypicking the gods with the absolute highest profitability and playing only those. The entry cost of 30 electrum is just too low compared to the possible winnings. Here's how my stats would look with skips:

  deck     Fast-Draw Ghostal   
  players     10 men   
  version    1.285 
  win-rate     38,4%   
  $ track ?    enabled 
  win-rate (n)     41,18%   
  games    500 
  Score/h     724   
  win-loss-(EM)     192-308-(153)   
  Score/h (n)    800   
  time (h:m:s)     17:00:00   
  FGei[c]* (http://elementscommunity.org/forum/index.php/topic,14626.0.html  7236   
  min/game     02:02   
  FGei[cn]* (http://elementscommunity.org/forum/index.php/topic,14626.0.html  7452   
      Statmasta™realtec                                                               

  score/h  FGei(c)   Statmasta™realtec    wins losses skips EM/Wins
   1351
   -21600
   -21600
   -21600
   -21600
   1358
   -21600
   1528
   1037
   1374
   -21600
   1089
   -21600
   1076   
   -21600
   1278
   -21600
   -21600
   1579
   -21600
   1388
   -21600
   735
   877
   1451
   -21600
   -21600
   -21600
   1509
   10767
   -21600
   -21600
   -21600
   -21600
   11032
   -21600
   7108
   7829
   7412
   -21600
   6948
   -21600
   8629   
   -21600
   7248
   -21600
   -21600
   7340
   -21600
   7701
   -21600
   7461
   6954
   8417
   -21600
   -21600
   -21600
   9971
Akebono
Chaos Lord
Dark Matter
Decay
Destiny
Divine Glory
Dream Catcher
Elidnis
Eternal Phoenix
Ferox
Fire Queen
Gemini
Graviton
Hecate
Hermes
Incarnate
Jezebel
Lionheart
Miracle
Morte
Neptune
Obliterator
Octane
Osiris
Paradox
Rainbow
Scorpio
Seism
Serket
   8
   
   
   
     
   10
   
   16
   17
   11
   
   20
   
   15
   
   18
   
   
   16
   
   9
   
   10
   13
   21
   
   
   
   8
    3
    20
    14
    15
    24 
    2
    17
   
    7
    1
    18
    5
    18
    3
    17
   
    16
    24
   
    12
   
    17
    8
    3
    1
    15
    26
    20
    2
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
     5
     
     
     0
     
     9
     
     15
     13
     11
     
     17
     
     4
     
     18
     
     
     16
     
     9
     
     5
     6
     20
     
     
     
     5


 

anything
blarg: