Suggestion to discourage cherry picking

If you think it might be a driver problem, see viewforum.php?f=79

Moderators: Site Moderators, FAHC Science Team

Post Reply
PaulTV
Posts: 179
Joined: Mon Jan 25, 2021 4:53 pm
Location: Netherlands

Suggestion to discourage cherry picking

Post by PaulTV »

Hi,

To start with: I love the fruit, this is more about cherry picking folding projects.

Today I've been looking at the results pages for several p17804 WUs.

https://apps.foldingathome.org/wu#proje ... 104&gen=19
https://apps.foldingathome.org/wu#proje ... 241&gen=35
https://apps.foldingathome.org/wu#proje ... =59&gen=51
https://apps.foldingathome.org/wu#proje ... e=4&gen=79

Those WUs are returned very often with Faulty 2 status. Looking at the user names, those F2 results seem to come from crypto folders, trying to game the system, cherry picking the projects with the highest PPD. That's hurting the science, seriously delaying the project, and it's annoying for the folders who just process the jobs offered by the assignment servers. Now, p17804 has significantly lower PPD than other projects, far lower than average, but there'll always be projects with the lowest PPD, so 'repairing' PPD for this project isn't a likely solution.

To discourage folders blocking WUs from certain projects, I would like to suggest changing the QRB calculation a bit. As far as I know the QRB currently looks at the success rate of a user/passkey combo over its entire lifetime, from the 11th returned result on. If it'd also look at the results of the last two weeks or so, avoiding certain projects this way will almost automatically hurt the QRB of those folders trying to game the system. Also, by looking at the last two weeks, next to the account lifetime, an honest spell of F2 returns won't hurt the account's QRB permanently.

So:

Unsuccessful = any job expired or returned as faulty or being dumped.
QRB_eligibility = (n_finished > 10) and (n_expired / n_finished <= .2) and not ((n_finished_last_2_weeks > 5) and (n_unsuccessful_last_2weeks / n_finished_last_2weeks > .2))

Of course, I'm just a simple folder, not knowing all the details of the stats and QRB system, so this idea may have some (for me) unexpected side effects, but I suppose something along those lines would discourage folders from cherry-picking.

By the way, I'm not opposed to crypto folding - I'm folding for the curecoin team myself, viewing it as monopoly-money that maybe at some point may return a little bit to support with the energy bills - but it should not delay the science of a below-average-PPD project.

Cheers,
Paul
Image

Ryzen 5800X / RTX 4090 / Windows 11
Ryzen 5600X / RTX 3070 Ti / Ubuntu 20.04
Ryzen 5600 / RTX 3060 Ti / Windows 11
NGruia
Posts: 13
Joined: Sun Apr 12, 2020 2:46 pm

Re: Suggestion to discourage cherry picking

Post by NGruia »

My problem is exactly the opposite of the one above. Regardless of the "cause" chosen, I receive over 70% WU for Cancer. Can server cherry-pick WU ? Example: system with 2 GPU cause Alzheimer. WU flow is something like this: 2 WU Cancer, 2 WU Cancer, 1 WU cancer + 1 WU Alzheimer, 1 WU Alzheimer + 1 WU Cancer , 2 WU Cancer and once a day (or twice in the best case) 2 WU Alzheimer. Is this OK ?
JimboPalmer
Posts: 2573
Joined: Mon Feb 16, 2009 4:12 am
Location: Greenwood MS USA

Re: Suggestion to discourage cherry picking

Post by JimboPalmer »

Welcome to Folding@Home!

Cause Preference is just a preference, the servers should give you a preferred WU if one exists, then any WU if any exist, then nothing if nothing exists.

If researchers have no WUs for your cause for your setup, the servers will try for any WU for your setup. There is no feedback method to encourage researchers to choose your cause.

(If I was going to discourage cherry picking, I would make the next WU lack the QRB)
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
aetch
Posts: 447
Joined: Thu Jun 25, 2020 3:04 pm
Location: Between chair and keyboard

Re: Suggestion to discourage cherry picking

Post by aetch »

If I was going to try to discourage cherry picking I'd give them a dirty great big red button on the control panel with a warning that if they dumped too many units in a row, or from the same project they would have to requalify for the QRB.

Make it easy to do, spell out the consequences.
Folding Rigs - None (25-Jun-2022)

ImageImage
gunnarre
Posts: 567
Joined: Sun May 24, 2020 7:23 pm
Location: Norway

Re: Suggestion to discourage cherry picking

Post by gunnarre »

Looking at some of those folding IDs, it looks like they have a very similar mix of CPUs, perhaps renting some kind of cloud folding service. Could it be that the service is unstable or automatically dumps low-PPD WUs? This is the bad side of the coin of rewards for folding - no pun intended.
Image
Online: GTX 1660 Super, GTX 1080, GTX 1050 Ti 4G OC, RX580 + occasional CPU folding in the cold.
Offline: Radeon HD 7770, GTX 960, GTX 950
Dayle Diamond
Posts: 27
Joined: Thu Jan 28, 2021 11:10 am

Re: Suggestion to discourage cherry picking

Post by Dayle Diamond »

I have a fast GPU paired with a first generation threadripper. It's got lots of cores but individually they don't go very fast.
So projects that require the two devices to check in regularly are gonna be a little slower.
My points per day fluctuated rapidly when I selected 'high priority' work.
I have found that picking Alzheimer's projects gets me project 18201 almost exclusively.
And that project has the highest points per day. It's not even close.

Is my cherrypicking a problem :?:
I want my contribution to be as efficient as possible to increase total system throughput.
But people here are talking about this as if it's bad behavior.
And the supply of 18201's seems endless. The project was announced in April and seven months later still has new work.
Neil-B
Posts: 2027
Joined: Sun Mar 22, 2020 5:52 pm
Hardware configuration: 1: 2x Xeon E5-2697v3@2.60GHz, 512GB DDR4 LRDIMM, SSD Raid, Win10 Ent 20H2, Quadro K420 1GB, FAH 7.6.21
2: Xeon E3-1505Mv5@2.80GHz, 32GB DDR4, NVME, Win10 Pro 20H2, Quadro M1000M 2GB, FAH 7.6.21 (actually have two of these)
3: i7-960@3.20GHz, 12GB DDR3, SSD, Win10 Pro 20H2, GTX 750Ti 2GB, GTX 1080Ti 11GB, FAH 7.6.21
Location: UK

Re: Suggestion to discourage cherry picking

Post by Neil-B »

Dayle Diamond wrote:Is my cherrypicking a problem :?:
Are you dumping/deliberately faulting WUs with low ppd? ... Are you manipulating a firewall to block low ppd wu servers? ... Have you written a script that in some way tampers with the automatic assignment of wus by the fah client? ... If so you are Chery Picking and most likely damaging the progress of science.

Have you stated a preference cause in one of the FaH Interfaces? ... Then you are not Cherry Picking ... The WUs assigned will if possible take that into account (but note it is only a preference) if there are available wus for that preference but this is by design the way FaH works and is not damaging to the progress of science ... There are other settings within the interface such as flags which can also impact the wus you get but again these are by design part of the FaH process and so again are not Cherry Picking.

At various times due to the availability of wus from various project and the specific kit being folded on it is quite possible and in no way unusual to have a whole series of the same project fold on a specific slot.

Tbbh FaH points are a very crude valueless method of indicating some level of contribution to FaH ... There might be a project that is so small it doesn't use fast/wide gpus very well at all - the benchmarking that will assign the BC to that project will give normal points for the slow/thin gpus but for the fast/wide ones low ppd will be delivered ... It is possible for the project to be limited to the slow/thin gpus but what if the project is really important and needs as much resource as it can get, especially the fast/wide gpus that whilst running a tad inefficiently will still be completing the wus much faster than the slow/thin ones? ... and what if that project is for the single specific cause that the donor wants to fold for - they may not care about the points and simply want to assist that specific science?

In all honesty points are a distraction - yes an important one in that many people feel the need to be rewarded for donations ... The important thing is that the researchers get supported and decisions as to assignment and wu allocation are managed within the FaH ecosystem in support of the science - not simply to ensure donors get the highest rewards.

Efforts are underway to improve Benchmarking and to improve Assignment but these things take time ... Internal Testing informing Researchers choices as to "size" of wus also tries to minimise some of the variations - but is the science either needs a very large atom count or a vary large number of steps then small/thin gpus may have to run for a number of days to complete these - timeout and expiration deadlines may be adjusted accordingly but some donor folders may find them "difficult" as they may need 24/7 folding on the slowest kit to meet deadlines ... For fast/wide gpus a very small atom count with a very high number of steps can mean lower than expected/hoped for ppd for extended periods of time.

If there were thousands of different types/styles of project and if there was absolute clarity on the capability of the resource being donated and the availability patterns of the kit then it might be possible to always allocate to a donor wus that meet the optimal configuration for their kit and availability ... but there aren't and there isn't so FaH actually uses quite a simple/naive assignment system which actually works for the main without wasting massive amounts of resource overmanaging this process ... but efforts are underway to try and improve this - it just takes time - not minutes/hours/days/weeks - maybe not even months - it really depends on the availability of developer resource (which is scarily scarce) and what else is happening in the world of FaH
2x Xeon E5-2697v3, 512GB DDR4 LRDIMM, SSD Raid, W10-Ent, Quadro K420
Xeon E3-1505Mv5, 32GB DDR4, NVME, W10-Pro, Quadro M1000M
i7-960, 12GB DDR3, SSD, W10-Pro, GTX1080Ti
i9-10850K, 64GB DDR4, NVME, W11-Pro, RTX3070

(Green/Bold = Active)
Mxyzptlk
Posts: 72
Joined: Wed Apr 08, 2020 8:55 pm
Hardware configuration: Lots... Look at my website: www.mxyzptlk.us
Location: California
Contact:

Re: Suggestion to discourage cherry picking

Post by Mxyzptlk »

Although I have never 'Cherry Picked', there is more than one way to 'game' the system for 'points'. I have been folding since March of 2020 and have nearly 6.5 billion points to date. I started with 2 average computers with 2700X's, one with an RX580 and the other a 1060. Since then I went crazy and have expanded ;). I never concerned myself which projects my systems were working on, just that they are working!
Maybe I'm being idealistic, but the only thing that I get (like most I suspect), is the feeling of hopefully doing something to help! Whenever I hear about someone else I know getting Cancer, I see my GPU's crunching away on Cancer WU's at a much lower PPD Average then normal and can't help but think that my GPU's hate Cancer too!
It is what it is. Do your part and everything else will work out.
I fold..... look at my folding setups here: www.mxyzptlk.us
Post Reply