Standardized Credits

[Zytozux]

Thanks for the response. I don't really have a problem with GPU's getttng more credit, because like you said, they do more work. I am mostly concerned about the wide credit range between projects like World Community Grid and Distributed Rainbow Tables. Getting 10k at DistRTgen takes 1 day, getting 10k at WCG takes 2 weeks. This chart is what I have an issue with mostly.

[Fire$torm]

Thanks for the response. I don't really have a problem with GPU's getttng more credit, because like you said, they do more work. I am mostly concerned about the wide credit range between projects like World Community Grid and Distributed Rainbow Tables. Getting 10k at DistRTgen takes 1 day, getting 10k at WCG takes 2 weeks. This chart is what I have an issue with mostly.

Yeah, I looked at that one awhile back. Its just that is no BOINC wide communication specifically between projects. Maxwell once told me that sort of thing is common in academia, which is where most projects originated from. Basically they are snobs and the only work that matters is the stuff each professor is working on. The exception would be related research and then only when they wish to share......

[zombie67]

Cross-project credit parity is impossible, for a number of well documented reasons. Adding together credits from different projects is like trying to sum toasters, miles per hour, and the color blue. The answer is meaningless.

Rather than adding credits from different project together for a total (meaningless), I choose to measure myself with MegaMilestones. That method makes the credits that a particular project awards meaningless. To compete in the MM game, means getting to a particular milestone (say 10k) in as many projects as possible. You want more of that MM than the next guy. Sure that means you will focus on the high paying projects at first, because those will be the quickest to complete. But after you achieve the MM in those, you will eventually have to move on to the lower paying projects too, in order to get more MMs. Just like your competition. So it doesn't matter if a particular project pays high or low. It will take each person at that project, the same amount of work to hit the MM.

Once you wrap your head around this, you quickly stop caring about the amount of credits that any particular project pays. It just doesn't matter any more.

Edit: Oh yeah, and the MM measuring method applies to teams as easily as individuals. Who is #1? It really depends on how you measure it.

[spingadus]

I'm with zombie on this. Total credits doesn't make sense as you cannot compare credits across different projects. This incongruity between the definition of 'being the best' and the false comparability of project credits has led to horrible attempts to normalize credit. An example is DA's creditnew, which in my mind makes the problem worse as it seems to penalize faster pc's over slower ones.

A better more accurate measure would be to compare overall average ranking of each person and team per project. You could easily give points based on percentile and then average them out over all active projects. This would make each project have equal competitive value regardless of how many credits they give. This could even be broken down into sub categories based on the type of project, for instance math vs bio-med. Or even best in nonactive projects.

It seems to me that the stats sites have the power to redefine the definition of 'top team' or 'top user'. As long as total credit is the basis for competition instead of average ranking we'll always have the rush to the high paying projects. It forces teams who want to be competitive to constantly crunch for the high paying projects at the expense of the other projects.

I wonder what Bok would have to say about this.

[zombie67]

A better more accurate measure would be to compare overall average ranking of each person and team per project. You could easily give points based on percentile and then average them out over all active projects. This would make each project have equal competitive value regardless of how many credits they give.

FWIW, this is similar to what FormulaBoinc does at the team level. They just reset the competition every year. It could be adapted to the user level easily enough.

[Bok]

FWIW, this is similar to what FormulaBoinc does at the team level. They just reset the competition every year. It could be adapted to the user level easily enough.

You mean something like this - http://stats.free-dc.org/stats.php?page=boincusersrank

I did this a long time ago, but there was not too much interest in it at the time, left it in the code though..

First column (rankpoints) is calculated per project as 10000 - ((rank - 1) * (10000 / total_num-users))

Second columnn is calculated as (user_score / total_project_score) * total_users in project

I don't recall all my decisions in coming up with this at the time as it was a few years ago.

[zombie67]

You mean something like this - http://stats.free-dc.org/stats.php?page=boincusersrank

Well, that is *obviously* the perfect way to score BOINC.

[spingadus]

You mean something like this - http://stats.free-dc.org/stats.php?page=boincusersrank

I did this a long time ago, but there was not too much interest in it at the time, left it in the code though..

First column (rankpoints) is calculated per project as 10000 - ((rank - 1) * (10000 / total_num-users))

Second columnn is calculated as (user_score / total_project_score) * total_users in project

I don't recall all my decisions in coming up with this at the time as it was a few years ago.

This is cool! It would be awesome to see more than 100 users and include teams as well.

Do you have any plans on continuing with it?

[spingadus]

FWIW, this is similar to what FormulaBoinc does at the team level. They just reset the competition every year. It could be adapted to the user level easily enough.

Nice. I like what they are doing, but I disagree with the points system. It should be more linear and include more than the top 10 ranks. It penalizes the teams that are more widely spread and gives more credence to those that focus on being in the top 10 in a limited number of projects. You could be ranked number 11 in 100 projects and get 0 points, while another team can be ranked 10 in only 10 projects and at least get 10 points.

[denim]

Cross-project credit parity is impossible, for a number of well documented reasons. Adding together credits from different projects is like trying to sum toasters, miles per hour, and the color blue. The answer is meaningless.

Rather than adding credits from different project together for a total (meaningless), I choose to measure myself with MegaMilestones. That method makes the credits that a particular project awards meaningless. To compete in the MM game, means getting to a particular milestone (say 10k) in as many projects as possible. You want more of that MM than the next guy. Sure that means you will focus on the high paying projects at first, because those will be the quickest to complete. But after you achieve the MM in those, you will eventually have to move on to the lower paying projects too, in order to get more MMs. Just like your competition. So it doesn't matter if a particular project pays high or low. It will take each person at that project, the same amount of work to hit the MM.

Once you wrap your head around this, you quickly stop caring about the amount of credits that any particular project pays. It just doesn't matter any more.

Edit: Oh yeah, and the MM measuring method applies to teams as easily as individuals. Who is #1? It really depends on how you measure it.

This has got to be the funniest thing I have read in a month.