Confusion about stuctures, generations and points

[[DPC]Mobster]

This is the case. I'm running the client and using Jeff's 2.2Beta3 because I'm really dumb :eek:

When I first start my XP2000+ it tells me that I will be generating some 115k structures a day. Not having participated in BETA testing I assumed :bang: that I would receive 115k of statspoints (which in Fase I were similar to structures).

The longer my client was running the farther my Structures Per Day dropped. Being your average statsaddict I sounded the alarmclock and started hitting unknowing people in my vicinity over the head with whatever I could get hold of :Pokes: :spank: :haddock:

After some carefull reading I kinda figured out the way it works, but I desperately need your confirmation.

My first generation consists of 10k structures and will award me with 50 points
My next generation consists of 50 structures and will award me with 50 x (1x1) = 50 points
My next generation consists of 50 structures and will award me with 50 x (2x2) = 200 points
.....
My next generation consists of 50 structures and will award me with 50 x (56x56) = 156800 points (being an XP1800+ running for 6 hours)

This last one seems highly unplausible to me. So this is when I loose control :looney:

Concluding. The number of structures shown in the GUI have absolutely no relevance to the stats showing here (http://www.distributedfolding.org/teampages/team81.html)?
And why don't I have 156800 points although the system is busy with generation 57...

Please help me.... :help: :cry:

[Digital Parasite]

dfGUI tells you how many structures/day you are currently getting. Generation 0 is a special case (it runs like phase I of DF) but the other generations will take much longer per structure and you will see dfGUI reporting much lower values once you get there. Since generations take varying amounts of time, even on the same machine there isn't really any way to predict how many points per day any machine will get.

The formula for points works like this:
- Gen 0 is worth 50 points.
- Gen x is worth 50 * sqrt(x).

So gen 0 is 10000 structures and you get 50 points.
Gen 1 is 50 structures and you get 50 * sqrt(1) = 50 points
Gen 2 is 50 structures and you get 50 * sqrt(2) = 70.71 points
Gen 3 is 50 structures and you get 50 * sqrt(3) = 86.6 points

and it goes on... The score is actually rounded so you don't get decimal values but you should see how it goes.

Jeff.

[[DPC]Mobster]

Thankx Jeff for clearing that up. Got my square roots and ^2 mixed up... :bang:

So is it possible that a certain generation will take less time to complete but will generate more 'statspoints'?

You say that there is no way to predict how many stats points a computer will receive, but is there something to predict then? Can a PII-600 yield more points than a PIV-2000?

How do we know our machines are working OK? And can we say I have 20GHz and you have only 10GHz so I will always be in front of you according to statspoints?

[Digital Parasite]

Originally posted by [DPC]Mobster
So is it possible that a certain generation will take less time to complete but will generate more 'statspoints'?

You say that there is no way to predict how many stats points a computer will receive, but is there something to predict then? Can a PII-600 yield more points than a PIV-2000?

How do we know our machines are working OK? And can we say I have 20GHz and you have only 10GHz so I will always be in front of you according to statspoints?

As you can see from the formula above, higher generations will always give you more points than lower generations. The speed of each generation depends on how often the protein gets "stuck" so it is possible that generation 5 could take say 30 minutes and generation 150 could take 10 minutes. In that case you will get a lot more points for crunching 10 minutes in generation 150 than doing 30 minutes in generation 3.

In general, faster machines will yield more points because they can try structures and re-try structures faster. But a fast computer that happens to get stuck a lot might generate less points than a slower computer for a particular generation.

Jeff.

[[DPC]Mobster]

OK. That is crystal clear.

So the more generations you do the more points you will receive BUT the points/time ratio doesn't have to increase with each completed generation...

[PCZ]

Why do they get "stuck"
I am monitoring some nodes with DCmonitor and they regularly get stuck. Usually between 5 - 20 mins. Fortunately they unstick themselves without any intervention on my part.

Brian

[[DPC]Mobster]

Mine has been stuck for over 2 hours now. Gonna hit the sack now (western Europe) and hope to have completed a few extra generations tomorrow morning... :(

[RaginSteveK]

an understatement..

I'm seeing as many as 140K iterations [ conformations] on a single step.. or "tight spot..", practically every step of the way, so it seems !!!!

how much extra credit am I getting for these machinations.. ?
[ couldnt possibly be enough ]

--
sorry--
tight spot thread covers this well...

[bwkaz]

Originally posted by PCZ
Why do they get "stuck" Because it's not random anymore. :)

I don't think you were around before the changeover (or did you start up the client on a machine or two? don't remember), but the phase 1 algorithm was basically "put amino acids onto the protein in a random place, and if any of them get too close to others, take them back off and re-choose the random place to put them". This algorithm did still get stuck sometimes, but not nearly as often (because it was completely random).

With the phase 2 client (and if you want a longer explanation, you can check some of the beta threads, down a couple pages), it takes the best structure from gen 0 (which is still randomly-generated), and tries to tweak it repeatedly to get it closer to a low energy (low energy is fairly well correlated with being close to the real structure).

This repeated tweaking has a MUCH higher tendency to get stuck by putting amino acids close together, because so many things aren't nearly as variable anymore. It was a regular occurrence during the beta.

What is nice, though, is that if one generation gets stuck, the laxness levels get brought up pretty high, which means that the next three or four generations fly by pretty fast. The laxness level is approximately "how far can I move AA's during tweaking", or at least that's my understanding of them.

[PCZ]

Thanks for the explanation.

I did run a few nodes over the last few days of Phase 1 but was never involved in the beta.

Seeing how often the PC's I monitor [The ones in my home] get stuck, It makes me worry about my unmonitored nodes.

If they stop I won't know.
Do they always unstick themselves ?

Also would I be correct in assuming that if the Phase 2 client is run on 2 identically configured PC's for the same length of time, that they wouldn't produce the same score.


Brian

[FoBoT]

Originally posted by PCZ
Why do they get "stuck"
I am monitoring some nodes with DCmonitor and they regularly get stuck. Usually between 5 - 20 mins. Fortunately they unstick themselves without any intervention on my part.



that is the way it is supposed to be, it means that your computer might be finding out some good data, instead of just pumping out masses of random/worthless structures

it is a Good Thing

[FoBoT]

Originally posted by PCZ

Also would I be correct in assuming that if the Phase 2 client is run on 2 identically configured PC's for the same length of time, that they wouldn't produce the same score.


Brian

yes, that is true

if you start a slower PC and a faster PC , with this new model, after 1 week of crunching, you might end of with similar OR radically different scores

it just depends on the path each starts down during generation "0"


(best i understand it)

[tpdooley]

We had systems stuck in a certain generation for over a day (my case) or 2 days (the record holder ;) for another Beta tester. Once it got done with the generation in question, it raced through the rest of them to gen 250. So don't worry about a few hours.. Just keep hoping that it's working on a really low RMS structure. :)

[[DPC]Mobster]

I'm trying to make a totally useless Excel-doc to determine how much points my offline systems have generated.

The only thing I want to know is what happens when ie I submit gen 2 (70,7106) and gen 3 (86.6025). Will I receive 156 points (both rounded down) or will I receive 157 points (first the sum and then rounded down) or will I receive 157.3131 points and it just shows 157 points in stats?

Question is really. Are the numbers really rounded up or down or does it just look like this?

[bwkaz]

Originally posted by PCZ
If they stop I won't know.
Do they always unstick themselves ? Most of the time. ;)

I haven't actually heard of them not unsticking, but some generations have taken quite a while on a couple of machines.

Of course, there's a payoff -- if the laxness levels get high enough, then the chances that the next generation(s) go much faster rise. Probability, true (it's not guaranteed), but it is nice to see when after a day of crunching one gen, the next five go by inside of an hour.

[Digital Parasite]

Originally posted by [DPC]Mobster
Question is really. Are the numbers really rounded up or down or does it just look like this?

The points are rounded down, so X.Y gives you X points.

Jeff.

[FoBoT]

if i did my math right, i think a complete 250 generation set will yield roughly 130,000 points

it appears that a complete 250 generation set with this protein takes (this is a WAG) 5-15 days depending on luck/cpu speed

[[DPC]Mobster]

Originally posted by Digital Parasite
The points are rounded down, so X.Y gives you X points.

Jeff. You probably have the feeling you've answered this one before, but I want to be sure because I'm making a sheet in which nonetters can fill in their current structures and their structures buffered and that then will generate their number of points buffered.

So

Sending in gen 2 = 70 points (70.71 rounded down)
Sending in gen 3 = 86 points (86.60 rounded down)
Sending in gen 4 = 100 points (100.00 rounded euh uh....)
Sending in gen 5 = 111 points (111.80 rounded down)

Sending in gen 2-5 = 367 points (summarized rounded down numbers) and not 369 points (summarized numbers and after that rounded down) (if latter is the case FoBoT than a complete set of 250 generations will yield 132196 points to be exact)

Are the points per generation rounded down by the client after completing each generation or are the points rounded down by ie the server after sending in the results?

Is this correct? TIA :thumbs:

[AMD_is_logical]

I've heard rumors that there is a bonus awarded when you finish a set of 250 generations. I've never heard how many points this bonus is, though.

[pfb]

Originally posted by AMD_is_logical
I've heard rumors that there is a bonus awarded when you finish a set of 250 generations. I've never heard how many points this bonus is, though.

I think that is just a rumour - I have had 2 250 sets now and haven't recieved any extra scores (none noticeable anyway)...

[Louis]

Yeah, there's a bonus alright.

"On your deathbed, you will receive total consciousness. So, I've got that going for me. Which is nice." Bill Murray, Caddyshack

But anyway, isn't the joy of finishing the reward? Just like beer, now you can start another round...

Louis
:harhar:

[shortfinal]

Originally posted by AMD_is_logical
I've heard rumors that there is a bonus awarded when you finish a set of 250 generations. I've never heard how many points this bonus is, though.

The plan was to give double points for generation 250 as a reward for producing all the generations.

So, I believe the equation is:
(50 * sqr(250))*2
which is: 1581

Shortfinal

[[DPC]Mobster]

It was the plan? They didn't go through with that one?

[MarcyDarcy]

Originally posted by [DPC]Mobster
You probably have the feeling you've answered this one before, but I want to be sure because I'm making a sheet in which nonetters can fill in their current structures and their structures buffered and that then will generate their number of points buffered.

So

Sending in gen 2 = 70 points (70.71 rounded down)
Sending in gen 3 = 86 points (86.60 rounded down)
Sending in gen 4 = 100 points (100.00 rounded euh uh....)
Sending in gen 5 = 111 points (111.80 rounded down)

Sending in gen 2-5 = 367 points (summarized rounded down numbers) and not 369 points (summarized numbers and after that rounded down) (if latter is the case FoBoT than a complete set of 250 generations will yield 132196 points to be exact)

Are the points per generation rounded down by the client after completing each generation or are the points rounded down by ie the server after sending in the results?

Is this correct? TIA :thumbs: Good question, didn't think about this one but also i want to know this. Howard/Brian can you please explain? :)

[shortfinal]

Originally posted by [DPC]Mobster
It was the plan? They didn't go through with that one?

I remember Howard saying he'd do this during the beta testing but I never verified it because I never got to generation 250. Howard? Is this bonus implemented?

Shortfinal

[Brian the Fist]

There is no doubling of the score for gen. 250 or a 'bonus', no. Scores are rounded down to the nearest integer just before being added to your total (so fractions will NOT cumulate).