Votes but no rating

[HotnessRater]

It should lose by a smallbut clear amount, since it already lost a 1-1 direct matchup, and you stated in another post that 1-vote matchups were best.

I never said that because one person voted on it, that it made it a valid match up. I never said 1-1 matchups were the best way to compare 2 pictures. I said that if you have a 100 votes, it is best to spread them out to 100 different matchups. Obviously individual 1-1 matchups would be less reliable but they would also have insignificant weighting because of the number of matchups you'd be able to have.

I didn't pick her at random, Wilde lost a 1-1 matchup to her

Again... a 1-1 matchup, which is irrelevant in the grand scheme of things.

I don't understand that. Were the other matchups in conflict? How do you decide which votes are useful and which ones aren't?

The other matches were thrown out because the competitor ended up not being in her range. If you have enough support to believe a picture is an 8, losing to a 9.5 means nothing. Also winning against a 3 means nothing. And losing to a 3 can be written off as a fraudulent vote as can winning against a 9.5. So if a picture does 50/50 against a large enough number of competitors between 8.5 and 7.5 I conclude they are an 8, even if they have beaten some 9s and lost to some 5s.

That is actually rather odd. The pool system I wnvisioned and described to you would peobably not see much change from initial placement, unless a pic showed signs of large numbers of erroneous votes. I mean, the pic itself doesn't change, so if the rating is done right there is no reason for the rating to change much, generally. An exception would be the top 100, which would be off limits to initial placement. Pics would have to battle there way up ro the top spot. It would give fans pics to root for, and to keep coming back to see how they are doing.

Yeah well, I can do more with less votes than your system

Hmmm. I think proper statistical methods will give apretty reliable number.

Prove it... write the algorithm. I think your method will fail spectacularly.

That is overstating things, no stastical method reaches 100% accuracy. The point of dividing into 3 decimal places is so you don't have an average of 6000 pics on average tied at any one rating value. 4 decimal places would reduce the average number of girls in each pool to 60, based on your 1.2 million current photos. Of course, as you add photosu , the a erage number at each rating value will go up.

It also predetermines how many will be in each group right? So even if I upload 1 million 9s, they won't be room for them all as a 9.

I need your database schema

No you don't. I'm not handing that out. You don't need to know how the black box works. Just define the interface that you need for your algorithm to work. I will give you a pool of matchups with at least x votes, whatever you want that to be. You can start with the BotD matches. That gives yous 1133 matchups right off the bat.

1. How many 1-1 pic matchups have at least 10 votes? How many have at least 7? How many have at least 5? How many matchups of 3 or 4 have all the votes going one way? The answers to these questions will show the feasibility of different starting confidence requirements.

1133 Battles of the day

There are at least 55,000 match ups with 5 or more votes. This number is only babesrater and only includes votes that had the same winner and loser. So it would skip over matches that are 4-1, 3-2, 2-3 or 4-1. I'll have to make a similar table for match ups and convert existing votes into that to get a more accurate count. It also doesn't include matches that could be used in a 5 vote matchup like 3-0, 3-1

If you make it 10 vote match ups, it will be at least

What are the daily, weekly, and monthly voting patterns? That is, how many votes do you get on an average Monday? And the other days of the week? Does it vary by time of the month? Month of the year? Histograms would be good. Specific answers to this group of questions would provide guidance in how quickly we can boost the number of votes on partially completed matchups to levels sufficient to provide somewhat reliable data points.

It's pretty steady. Don't worry about how quickly votes come in. I will give you matchups with X number of votes. That could be 50000 in a day, it could be 0. Either way your algorithm should be able rank them.

The goal of these initial assessments is to gain some real numbers about how many pics would have actual ratings after an initial population of rating pools, and provide an estimate of how fast we can get additional pics out of the unrated state and into initial rating pools.

Again, that isn't really your concern. Just worry about correctly ranking what I can give you. If your match size is 5, expect

my dad died yesterday morning.

Sorry for you loss

 

[HotnessRater]

What do you consider to be a "qualified matchup"?

 

[HotnessRater]

Well, so your system is scoring this pic as 7.52, with 122 votes:

https://hotnessrater.com/full-sized-picture/2998502/gabriella-wilde

If that rating is accurate, then in a Battle of the Day it should score near 50% but be a clear loser to this pic:

https://hotnessrater.com/full-sized-picture/49837/taylor-chapman

Gabriella Wilde is up to a 7.74 and Taylor Chapman is down to a 7.27

 

[HotnessRater]

All the more reason to judge Wilde's loss to Chapman as erroneous data. Yet your system cannot do that, its data is settled, Wilde lost, and it will use that data to hold down Wilde's rating.

Again, I'm not saying Wilde lost. Why are you have a problem understanding that it just a single vote (and one vote that is no longer even used in either of their ratings because it has been discovered to be unreliable)

In a multivote rating system, Wilde would get enough votes to overcome the negative vote, and the matchup would become the victory it should be. That is the point I have been making, and you keep trying to deny it.

My point all along is that that vote we are discussing has been disqualified from affecting either pictures vote. That vote is not holding Wilde's rating down. Because of their spread in ratings, that vote has been deemed unreliable... As far as the calculation is concerned, it never happened. So I was able to disqualify that vote with zero additional votes. That is my whole point to you that you ignore. I don't need 5 or 10 votes on a single match up to realize that it is invalid. I can do that by comparing the other results of each picture.

 

[HotnessRater]

All the more reason to judge Wilde's loss to Chapman as erroneous data. Yet your system cannot do that, its data is settled, Wilde lost

Yes and my system realized that. Yes it CAN do that. The data was not settled, it realized the vote was unreliable and was thrown out... and it did it without extra votes between the two

 

[HotnessRater]

If A beats B, C, D, E, F and G
and B beats C, D, E, F and G
and C beats D, E, F and G
and D beats E, F and G
and E beats F
and F beat G

but then G beats A, my software will throw out that vote. It will remember it and report it in the win lost list, but it won't use it for calculations of A,B,C, probably D and maybe even E and F depending on other votes and what the spreads are.

Now if G somehow goes on a winning streak and beats H,I,J,K,L,M who are rated the same as A,B,C,D,E,F respectively, then it goes back into unrated category. At this point it is probably rated (but reported as unrated) around the same rating that C or D is. So I throw more Cs and Ds at it until it rises or falls or maintains close to a 50/50 win loss ration with the C and D crowd. This is why you can see some pictures with lots of votes without a rating. They jump around between groups until there is enough data within that group that they belong there... and the votes outside that grouping are not considered in their rating.

 

[HotnessRater]

Well, if the A through G matches are all supported by enough votes to have very high confidence values, then G isn't going to beat A

And that is where you are wrong. Often you are going to run into Rock/Paper/Scissors pictures that are going to throw your whole system out of wack

Of course, if all the pictures are equal, then a 1 vote match is 50/50, and just like a coin flip, you can get a streak where everything runs the same way. Or maybe the voter just always votes for the left-hand pic.

There is no such thing as a left-handed picture. Their placement (right or left) is randomized after the matchup is determined

Well, if the A through G matches are all supported by enough votes to have very high confidence values, then G isn't going to beat A

And my point is I still determined that it was a bogus vote without having to waste all the votes building "high confidence values" and was able to use those votes to bring in H-Z and many more which get their confidence values from being compared to a large number of pictures and a huge web of interconnectivity instead of just a handful of "high confidence" match ups

 

[HotnessRater]

As for the rock/paper/scissors problem, it will only come up in two places: initial population of the pools from your existing data, and in the top 100 list.

I think you are wrong

In most professional leagues, division champ is determined by overall record, not head to head matchups.

Which is exactly how it works now.... and is why I don't care much about individual match ups. I would rather every team play every other team in the league once than have each team only play 10% of the league over and over

If push comes to shove, resolve it by hand by deleting a problem vote, but that is a last resort.

Haha... that's not going to happen. I'm not going to resolve thousand of matches by hand

 

[HotnessRater]

What is the average win percentage of battle of the days?

No idea

 

[HotnessRater]

Isn't that something you can run a query for?

Not easily. I probably could figure it out if I wanted to. I'm not really sure why you would care though

 

[HotnessRater]

Not easily. I probably could figure it out if I wanted to. I'm not really sure why you would care though

From what I can tell its about 66%

 

[HotnessRater]

Well, so about 34% of votes were on the losing side. 1300+ battles is enough to be statistically significant. System-wide, probably 30-40% of your 1-1 matchups have been decided by the minority opinion vote, which happened to get the matchup assigned to them by luck. In other words, in all likelihood, around 35% of your vote data is wrong. This is what your ratings are built on.

I don't agree with that conclusion. Just because most battles of the day were decided by a 2:1 ratio doesn't make the 1 vote 'wrong'. He just had a different preference.

AND FOR THE MILLIONTH TIME, 1-1 matchups, aren't "decided" any more than a single NBA game decides who the number one team is in any given year.

 

[HotnessRater]

I'm not explaining this again.

 

[HotnessRater]

No it's not like deciding the game on who scores first. I don't decide anything on one vote. I decide a pictures score based on many votes.

It's like deciding who wins the season based on all their games. If I made each matchup multiple votes, it would be like crowning a champion of a season after playing 3 games when most of the teams haven't played each other.

But you are just simply wrong about the validity of your rating system. I look at thousands of pics on your system each day, and a lot of the rating numbers make me grimace.

You haven't been able to show me a rating that is out of line yet....

I told you to design a better algorithm if you think you are so smart but I have yet to see one.

 

[HotnessRater]

That should be enough votes to both give each matchup enough votes and to deal with your current unrated pic backlog.

I have half a million pictures that haven't even been approved yet, nevermind getting them rated.

Still waiting for you to show me a picture that has a bad rating.... until then, all your whining and bitching about my algorithm (which you have no idea how it works) is invalid.

 

[HotnessRater]

Does your system keep historical info on ratings? That is, can you graph an image's rating changes over time?

No

 

[HotnessRater]

I've shown you a number of pics and matchup results that were out of whack. Saying that I haven't done so is just not true.

That is not true. You showed me single votes you thought were out of wack but when we looked at the ratings, everything seemed fine. Show me again if you think I am wrong.

As far as the bet goes, I'm not doing a bet for $1... it would be more hassle to try and pay it than it would be worth.