manga04_jpg login_tab_left_jpg
Username:   Password:    Forgot Password?
App
Try out our new iPhone application!
App
Manga Poll
For a really long, ongoing series, you prefer to
Keep reading since you're already so invested
Keep reading as long it's good
After a certain point, just stop and wait for it to finish then read the entire thing at once
Just give up reading after a certain point
 
See Old Polls

Manga is the Japanese equivalent of comics
with a unique style and following. Join the revolution! Read some manga today!

Coded in ConTEXT

Join #baka-updates @irc.irchighway.net

RSS Feed
 
center_left_tab Forums center_right_tab

You are now viewing a topic.

Bayesian Average switch button

Back to Suggestions & Bugs


You must be registered to post!
From User Message Body
Uruner
Post #282740
user avatar
Member

8:01 am, Apr 29 2009
Posts: 48


I think that there should be option to turn of Bayesian Average in your control panel if you want to.

I agree that it's best to have it switched on most of the time, but when you've read everyting that interested you from popular manga, becouse of Bayesian Average it's almost impossible to do proper search trough the ones that are rarely viewed.

Trough Bayesian average correctly reduces the effect of a single anomalously large or smal vote- it also makes sure that every manga with less votes gets allmost the same rating.... making it veery dificult to find trough search engine.

The too high and too low ratings are sometimes the best way to find interesting, unpopular manga.

That's why i would like to ask if it was possible to add this option to User Control Panel



________________
Cradle of Monsters
Destroy and Revolution
The Meteor
robbit
Post #283012
user avatar
I am the Devil
 Member

1:41 am, Apr 30 2009
Posts: 2083


whats the Bayesian Average?

lynira
Post #283019 - Reply to (#283012) by robbit
user avatar
Site Admin

2:02 am, Apr 30 2009
Posts: 433


Quote from robbit
whats the Bayesian Average?

You know how you can rate a manga, yes? There is an actual average and a Bayesian average.
(It will be something like this on the series profile:
"User Rating
Average: 9 / 10.0 (1 votes)
Bayesian Average: 7.69 / 10.0")
It's basically a more "fair" rating, because it acts to "[reduce] the effect of a single anomalously large or [small] vote," like Uruner said. It's weighted in some way so that it can do so... According to wikipedia, "a Bayesian average is a method of calculating the mean of a data set where there is a known prior probability of the value being estimated. It is of particular value when calculating means of multiple differently sized data sets from a larger population." Research it yourself if you want to know more.

As for the suggestion, you mean you want to be able to see (and sort by) the actual rating on the list of series, without having to click on each one? Hm... personally, I can't decide whether it would be worth the trouble of implementing or not... not that I'm one to complain about extra options.
Anyway, if unpopular manga is what you want, you can look for series that are on the lowest number of lists using the "series stats" section.

Last edited by lynira at 2:23 am, Apr 30

________________
Manga Cover Database
Uruner
Post #283128
user avatar
Member

8:15 am, Apr 30 2009
Posts: 48


It isn't just to search trough most unpopular manga.

Series Status have all sort of things mixed up in it, if you look at it that way, and you can't sort anything with it (like genre). It's no search engine.

My point is - to be able to search trough "middle popularity" where some people given their votes - and not series where there is no any solid info .
It would allow a lot wider style of search.

Besides implementing it shouldn't be such a hard thing...
Just changing from "Bayesian" variable to "normal avarage" variable in several places of code and it's done. It's not like you change search engine or style of page.

________________
Cradle of Monsters
Destroy and Revolution
The Meteor
pnyxtr
Post #283190
user avatar
Member

2:31 pm, Apr 30 2009
Posts: 142


Another possibility would be to tweak the C value a bit in the Bayesian average. C, of course, being the number of votes required before real votes start to get more important than the fictitiuos average. (It looks like it's around 30 or so, in my highly unscientific study.)

Which is fine, though. I would probably have set it to maybe half that, but it's very much a matter of personal opinion, and about what kind of safety you want against people trying to rig the vote. I'm just saying, it's a possibility. And even easier to implement.


silencer
Post #283227
user avatar
The last Blood Elf
Member

5:47 pm, Apr 30 2009
Posts: 200


Do you mean C is the number of votes proportional to a credible average value? lol i just get to know about this stuff, due to simple calculation C=30 , also i'm wondering about the actual formula they used in this system? it's not the one in wiki since the prior mean value would be different for each manga eek and i only use that formula for manga that have similar distribution of votes to get the value of C.

Anyway, if there's a modification, then just set C=0 for the switch off option.

________________
User Posted Image
pnyxtr
Post #283352
user avatar
Member

11:35 pm, Apr 30 2009
Posts: 142


It's not quite that. I know quite a bit more than I think is good to let on, but the assumed average does exist, for sure, and is probably quite close to 7.62. (I have to wonder about this, since 7.62 is also .30 cal.)

And C isn't 30 (or maybe it is), I'm just saying it's somewhere around there.


silencer
Post #283465
user avatar
The last Blood Elf
Member

4:51 am, May 1 2009
Posts: 200


Alas, my point is, i actually found C = 30 (approximately) for some manga that have similar ratio between different ratings, since the prior mean values (m) are most likely to be the same between them ( i know they are supposed to be the same between every manga), then i substitute C=30 into the formula, found m = 7.71 - 7.61 for some manga with below 900 votes, so i think 30 is an acceptable solution, but after that for Naruto( 1868 votes) m = 7.37,Bleach(1669 votes) m= 7.83, Vampire knight(1494 votes) m = 7.58, Claymore (1250 votes) m= 7.82 and if you assume that these manga have the same m you'd find ridiculous value of C...

So, i suppose that the simple formula in wiki is most applicable for medium to low number of votes, and the actual formula is a formula that can be approximately reduced to that formula with small value of n.

Edit: oh,and consequentially, the value of C in the actual formula can be very different from that of simple calculation.What i did is just rough estimation.

Last edited by silencer at 5:33 am, May 1

________________
User Posted Image
pnyxtr
Post #283768
user avatar
Member

10:59 am, May 2 2009
Posts: 142


If you look a bit closer at the most popular series, I think you too will come to the conclusion that for series with many votes, what's presented under Bayesian average is just the plain average rounded a little less. Exactly how many votes is best left up to the imagination, I think. This makes perfect sense, of course, there's little reason to use a Bayesian average on a very large sample. (But it does skew the comparison a tiny bit in favour of super popular series.)

And if you try one series with just a single vote of 7 and one with a single vote of 8, you can easily see that m has to be somewhere between 7.62 and 7.66.

What I meant in the first post was that until you get up to C number of votes, the assumed average will be more important to the end result. At C, the real votes will be just as important as the assumed ones. And at C+1 and up, the real votes will be more important.


silencer
Post #283810 - Reply to (#283768) by pnyxtr
user avatar
The last Blood Elf
Member

1:40 pm, May 2 2009
Posts: 200


Quote from pnyxtr
If you look a bit closer at the most popular series, I think you too will come to the conclusion that for series with many votes, what's presented under Bayesian average is just the plain average rounded a little less.

Yeah, it's obvious since increasing n (the number of votes) or decreasing C both make Bayesian average (BA) nearer to the arithmetic average, but i don't care about it.
Quote from pnyxtr
What I meant in the first post was that until you get up to C number of votes, the assumed average will be more important to the end result. At C, the real votes will be just as important as the assumed ones. And at C+1 and up, the real votes will be more important.
That's right, but what i was wondering about is the possibility of another formula used by the system, and now i start to think the different values of m is probably just the problem of error in calculation laugh After taking calculus, i found that the error of m depends on the error of the BA value with an amplification factor equal to (n/C+1), lol, which means it can be up to 0.6 if the error of BA is 0.01and n=1800 (actually i left out an insignificant term).
Also, the error of C has a wider range, lol so it can be weird in some cases. End of the off-topic query biggrin

________________
User Posted Image
Uruner
Post #586931
user avatar
Member

8:24 pm, Feb 10 2013
Posts: 48


I re-new my suggestion after 4 years.

________________
Cradle of Monsters
Destroy and Revolution
The Meteor
You must be registered to post!

Back to Suggestions & Bugs  Back to Top

Search This Topic:
 
Manga Search
MANGA Fu
MEMBERS
TEAM-BU


footer