Rating changes

[HotnessRater]

There are two problems with this. First, other pics are having their own ratings inflated by defeating these overranked pictures, perpetuating overrating as new pics come in.

This shows you have no idea how my ranking system works. If a picture drops the pictures that beat it no longer get their ratings 'inflated' by that picture

 

[HotnessRater]

Note that with this system, it would be very rare for a pic to end up a 10.0 with 15 individual votes, like happens comparatively frequently atm

Yes because nobody would have a rating after just 3 matchups...

I could just increase the number of votes to 25 before it becomes unranked and then you wouldn't see 10s anymore.

A major flaw with your idea is the assumption that just because PicA beat PicB and PicB beat PicC that PicA would automatically beat PicC. It doesn't work that way. Definitely not in just a 5 point match. So when PicC beats PicA how do you rank them? And thats just with 3 pictures. When you have 50,000 contradicting matches how do you rank the system?

 

[HotnessRater]

Also you have a finite number of placements at each point. So if a picture is legitimately a 9.0 but you have enough 9-10s to fill all the spots for 8-10, that picture gets pushed down below 8. The more pictures you get, the lower almost all pictures will score.

 

[HotnessRater]

If a users picture is a 7.5, it can't drop to a 6 just because people uploaded a ton of supermodels. That user is still legitimately a 7.5

 

[HotnessRater]

I will however adjust when pictures are considered 'rated' It used to be higher because and I made it easier for them to be rated but I think I went too far which resulted in pictures showing ratings before it was fleshed out enough

 

[HotnessRater]

So, great, you just saved the taxpayers loads of timeand money. From now on, all elections will be decided by one vote. First voter decides, the rest just make problems so send them home. The more voters, the worse the problems. Election returns take 2 minutes, and are all done by noon. We can have regular programming instead of listening to "experts" blather until 2 am. No more recounts or hanging chads. Hmm, I really think you've got something here, good job.

Oh, and let's fire every statistics professor and numerical analyst in the country, and have them go do something useful. All the people and companies who spend billions of dollars every year are just wasting it, it is all make-work. None Of them are needed, because sample sizes of one are the best way to be as accurate as possible. Simplistic solutions are always the best, nothing can be as accurate.

Flip a coin once, and you know what you will get if you flip it each succeeding time. Get one person's opinion which babe is hottest, and there is no chance they are an oddball, and most everyone else will disagree with them, fouling up all the decisions you build on top of that first one.

The thing you aren't getting through your head is nothing is decided by 1 vote. It is the accumulation of all that pictures wins and loses. Their is nothing statistically unsound about that.

 

[HotnessRater]

Obviously, you have to compromise one way or another. You can allow loads of pics to clutter up your rating points at the top of the scale, as you currently do. If you do that, the numbers lose their significance and their punch. "What, another 10.0? That's the third one I've seen in the last 5 minutes, and none of them are as good as that 9.3 with 2000 ratings. Weird."

So this can be fixed more by just changing the number of votes it takes to get a picture rated. If you rate them after 1 vote and they win, they are a 10. Obviously that isn't ideal. Waiting until 2000 votes doesn't work either. While being a lot more accurate, it would mean that pictures would take a longer to get rated and there would be a lot less rated pictures. I don't need to totally change the way the rating system works. I just have to find a balance there. I have recently increased the number of votes it takes to get rated and you are probably seeing less 10s.

 

[HotnessRater]

Once the limit is reached, a new pic must bump a pic out to make room. Otherwise, the pics you are rating really aren't better, which is what rating them should mean. Allowing lots of pics at the very top is just corruption of any sustem of comparisons.

I will explain this again. The 1-10 rating system is comparing people in general. Models are models because they are in that top 0.01% of the population. Therefore it makes sense they all fall into the 9+ category. Actresses are generally falling into the 8+ category. The hottest of the hot supermodels are going to rank close to 10.

If there is an issue with girls not being rated "correctly" it is more in the number of votes required to get a rating than it is in revamping the entire system. (and the multiple vote per match up system would require a lot more votes to get them accurately rated)

 

[HotnessRater]

Am I correct in my assumption that lower numbers mean pics have been on the system longer? If so, are exceptions made? Are new pics sometimes inserted at lower numbers? Can you provide a rough table of when various number ranges were added to the system?

Yes, No, No, and No (and why would you want that?)

 

[HotnessRater]

Well, they certainly won't all contradict. The more votes you have for a particular matchup, the more reliable the outcome is. That is, the more you can feel secure that the outcome is correct, and not just one person's honest opinion (which is subjective, and YES, all votes are subjective) at best, and off-the-wall goofiness at worst.

When you have a loop as you describe, which could (and most likely will, given the huge number of pics you have and the geometrically larger number of potential matchups) happen, you can do several things, or a combination of them. In no significant order:

A) assign more votes to be taken to raise or lower confidence, to resolve the situation by establishing which matchup is incorrect. If two of the three are strong, you can guess that the third is an anomaly, and concentrate extra votes there.

B) you can check their records against opponents in common, to see if a pattern emerges.

C) if the matches were all close, you can guess that the pix are a dead heat.

If I wasn't clear before, the tree is transitory. It is a way to establish an initial sort order. Then pix are assigned into your rating pools. There is one pool for each possible pic rating. Once the pools are established, the tree goes away. So, with the loop above, the three (or 4 or 5 or whatever pics) can just be placed in the same starting pool. After that, they will move up or down based on whatever new data comes in, as long as confidence in matchups is kept high. Only make solid decisions on strong data, that is how you maintain confidence that your ratings are correct.

Be clear that there will be a LOT of pics in each rating pool, I've discussed this elsewhere, dividing your 1.2 million pics amongst a limited number of possible ratings.

Pics within a pool are considered tied, until one or more either shows inferiority to the pool, or shows up as outstanding. Then they move one or more pool locations based on how strong or weak their records. They will get a new group to be tested against, to see if they have reached their more or less final resting place (they may creep up or down a little, but when you get their number right they will tend to stay put, running around 50% in future pool matches, appropriately higher or lower out of pool.

You are making this way, way, way more complicated than it needs to be.

If you think you can do better than what I am doing, write the code. Make a rating page and I will host it alongside our page. Tell me what methods you need for an API and I will provide them for you. We can have the HotnessRater system and the Morgan Rating System. Let's see which one works better.

 

[HotnessRater]

So, each time a rating is recalculated, it uses the new value for pics it as matched up against? That makes sense. But, when it is recalculated, do you then go and recalculate the ratings of all the pics who used its previous rating when their ratings were last calculated?

Yes, I have services that are constantly recalculating pictures, and then if that pictures rating changes, it queues up all pictures that have ever went up against that picture to have their ratings recalculated.

With my pool system, you don't need to calculate ratings at all, a photo has the rating of its current pool. When the pic moves to a different pool because it either beat out or lost to a large chunk of its poolmates, then it gets a new rating value, and a new set of matchups are created to be doled out to raters.

But it is flawed in many others ways that makes it completely infeasible.

You are probably running some version of LINUX, with mysql?

Nope, we are Windows and .net

Email me your current database schema - all tables, field names, types, and sizes, primary and secondary indices, current number of records in each, and rable relationship diagrams. I can design additional tables to support the pool system, and provide you with a detailed algorithm for initial population of the tree I described and placing the data into the pools.

Well I don't give away our DB schema to anyone for a number of reasons, least of all security. I can design tables and implement the pool system... but I won't because from my experience in doing this for years, I don't believe it would work well. I think predefined pool sizes are problematic, especially since we have no way of knowing how many should be in each pool and new pictures get added all the time. I also do not think it works well over a 1-10 range. It also requires orders of magnitude more votes to get a picture rated... and when it all failed you would just accuse me of not coding it right.

I see so many holes in that system, it is a complete no starter for me.

The problem isn't coming up with an overall design for what you proposed. The problem is actually writing the code to make it work and how you would deal with all the gotchas that I can already see clearly as day from my experience in both writing several iterations of this rating system and 25+ years developing software professionally in general

 

[HotnessRater]

Start with a normal bell curve

Because the pictures are not a random cross section of society but instead pictures that are custom chosen to be on the high end of the spectrum, a standard bell curve no longer makes sense. It would be like taking a random sample, picking out only the best 10% and then trying to apply a bell curve to those. It makes no sense.

That is an overly-dramatic exaggeration.

I don't think it is.

Does your code currently allow you to manually specify a start point for rating, when pics are uploaded and viewed for moderator functions? That is, can you tell the system "start this pic out around 8 (or 9 or whatever)"?

Not really, they all start at the same number.

I suspect pics may have more votes than the system tells me,

No we tell you all the votes a picture has. We just don't necessarily use them all in the computation.

I'd give the moderator the ability to say "around 8.5" or whatever

So if I could make that judgement and type it in and have that process take 1 second, it would have added 57.8 days of moderation time. No thanks, moderation is slow enough as it is already. I don't need to burn an extra half hour to an hour a day entering ranking that the system is supposed to figure out anyway.

Shouldn't take more than 45 votes if the initial guess by the moderator was a good one, and we are using 5 vote matches to start. Yeah, that is more than your system, but not orders of magnitude more

It is 9 times more than what the system was a week ago. 4.5 times what we require now. That's the minimum anyway, assuming we use all of them in the calculation. In practice, it is probably 2-3 times as many votes. It's unacceptable.

Personally, I'd cut the number of pics by 20-40%

I know you would. I wouldn't. I'm not going to do that to my uploaders.

I'm not sure why you keep talking about this pool system. It isn't going to happen. Ever.

 

[HotnessRater]

Okay, here is a good example of the sort of thing your 1-vote outcomes permits:

https://hotnessrater.com/picture/1185440/anna-prugova

So, looking at the "wins", it appears to me that 12 or 13 of the 20 were joke votes. This has skewed the rating far higher than it ought to be. Any other pic that beats it gets overrated because this pic is overrated. Any pic that loses to it isn't treated correctly, either. A pic that loses a joke vote gets their rating dragged down undeservedly. Pics that actually deserved to lose aren't dragged down enough.

60-65% error rate is pretty bad. Then that error infects the ratings on other pics, which spread it to other pics.

Except you don't understand what I do with the votes, which ones I use in the calculation... and which ones I don't

8.67 - she is a pretty girl. I don't think this rating is out of line

Look at the others against her that fell into the 8.69-8.65 range. I think they are all pretty close. I would actually use this as an example of how the system placed her correctly even if all the votes weren't correct.

 

[HotnessRater]

This pic had its rating dragged down in part because of a joke loss to the above pic.

https://hotnessrater.com/picture/4062366/asuka-kirara

Nah, she just isn't all that attractive. No hips, no boobs. If anything, she is rated too high

 

[HotnessRater]

Do you specify the battle of the day? Do you take requests? I'd like to see the outcome of these two pics:

https://hotnessrater.com/picture/4400609/yixin-tang
https://hotnessrater.com/picture/4062366/asuka-kirara


The kirara pic lost a 1-1 matchup, but it seems too hotnfor that to me. A battle outcome would tell me a lot about your voters' opinions compare to my own.

Sure I can do that

 

[HotnessRater]

https://hotnessrater.com/battle-of-the-day/1121/yixin-tang-vs-asuka-kirara

 

[HotnessRater]

Did you say she has no boobs seriously?

Well in that other picture, you can't really tell

I was just commenting on the two pictures you posted... since I don't really know either of them

 

[HotnessRater]

So, how many voted in the battle? Was it exactly even, or did they round off to 50% each?

So far it has been 52 votes. (51%-49%)

Not really the outcome I expected, but it does show that one vote cannot be relied on to determine the outcome accurately. Your system is treating the match as if there was a clear victor, and adjusting ratings because of it, when it was a dead heat

No the system does not treat it like there was a clear victor. I'm not sure where you are getting that from. It just shows that it is 51%-49%... Adjusting ratings because of it? What are you talking about? The battle of the day ratings have nothing to do with their rating.

 

[HotnessRater]

Another rating injustice:

Another? You haven't shown me a single instance of rating injustice yet.

https://hotnessrater.com/full-sized-picture/4505557/dianna-dahlgren

Defeated

https://hotnessrater.com/full-sized-picture/4865197/shir-elmaliach

Seriously?

See you still don't get it. By showing the individual votes, we aren't declaring that picture is better. We are just showing the individual votes, which may or may not be right. I'm not claiming those individual votes were accurate. It simply won an individual vote.

Dianna Dahlgren currently has a 8.7 range. Shir isn't even rated yet! You can't claim a rating injustice when the picture isn't rated yet... After they are both rated, if Dianna's rating comes out higher, then you have a legitimate claim for a 'rating injustice' - but just because someone made that vote doesn't mean anything.

The pic of Shir also "lost" to this one:
https://hotnessrater.com/full-sized-picture/3985864/luna-sobrino

Again... a single vote. It means nothing.

I looked up the rating of the Shir picture and although it still needs more votes, it is currently higher than the other 2 pictures in question... so it seems like the system is in agreement with you. I'm not sure what the complaint is.

Single vote matchups leads to a lot of junk,results, it's like deciding a basketball game by who scores first.

Not at all... it is like listing the people on both sides that scored baskets. Again, the win/loss pictures aren't meant to be a definitive declaration of that picture being better. It is simply showing the voting process. To find the winner, you have to compare their ratings.

BTW, when did the Kunis pic get added to the system?
https://hotnessrater.com/full-sized-picture/5700774/mila-kunis

That picture has been in their forever (or at least a lesser quality duplicate was there). I think it was recently replaced with a better copy

 

[HotnessRater]

Mila Kunis won because she's Mila Kunis, the pic itself Is preetty good, but has 3 lifetime views.

Lifetime views mean very little. If a picture is newly uploaded, it has 0 lifetime views.

Pictures with a high number of views, could just mean that, for whatever reason, Google has given that picture a higher place in its index.