This is just a tooting-my-own-horn progress update with a program for plotting data from job_log_einstein.phys.uwm.edu.txt. The basic functions are written to plot task times or search frequencies by E@H Project.
Cecht, you're doing a great job at making a graph to plot the frequency and tasks/day, and possibly others. And that is all well and good. But, not to rain on your parade, why do this? To what extent is this going to be useful to anyone within the BOINC community, or to be more specific, Einstein@home?
Unless you're getting your toes wet before trying to tackle another BOINC project, or some other tasks like another program, I don't quite see why this particular project is important. I realize that monitoring the project frequency and number of tasks/day may be good for you. I just am having trouble understanding the why, or to what extent.
It’s interesting to see the frequency plotted for each series.
but maybe something got misinterpreted when I suggested looking at the frequencies. I wanted to see avg runtime for the frequency ranges. Not just the frequencies themselves. Which means grouping the task results by frequency then plotting the runtimes for each frequency range.
To what extent is this going to be useful to anyone within the BOINC community, or to be more specific, Einstein@home? ... I just am having trouble understanding the why, or to what extent.
Fair enough. The original idea of the counting program was use task completion rate as a measure of productivity because completion rate is not easily obtained from BoincManager or a user's account page. I've used it to compare hosts, Projects and GPU settings and thought others might also want to do that. Also, i get a kick knowing how many tasks my hosts can grind through in a day, month, or year.
The plotting program I'm developing is a E@H data exploration tool. At the moment, I'm not sure what usefulness it might have.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
I've used it to compare hosts, Projects and GPU settings and thought others might also want to do that. Also, i get a kick knowing how many tasks my hosts can grind through in a day, month, or year.
The plotting program I'm developing is a E@H data exploration tool. At the moment, I'm not sure what usefulness it might have.
Ahh... I now am beginning to see some of the benefits of it, to an extent.
If you could, can you retrieve the data results from differing hosts? That is to say, it would be useful to compare different hosts as far as productivity, or number of tasks each host grinds through per day/month/year for the same project.
If the plotting program were to be useful (at least to me), I would think if it had an option to graph out the results over time as opposed to plotting dots, and also comparing different hosts, then I could see the benefit.
I've never seen the need to graph anything out or install anything extra. The tasks per day/per month for each host is available for each project I run through the simple BoincTasks interface.
It’s interesting to see the frequency plotted for each series.
but maybe something got misinterpreted when I suggested looking at the frequencies. I wanted to see avg runtime for the frequency ranges. Not just the frequencies themselves. Which means grouping the task results by frequency then plotting the runtimes for each frequency range.
I re-jigged the frequency plots to run-time vs. frequency for FGRPG1 (N = 139744) and GW O3 (N = 119607) tasks. In short, except perhaps for limited frequency ranges or data series, there is no correlation. Spearman's correlation for FGRPG1 data is -0.63 and for GW O3 is -0.44.
I think the plotting program is best for data exploration (see screenshots below). If you or anyone can think of some E@H plotting or reporting feature that might be useful, I'll be glad (?) to give it a shot.
The poor or absent correlations are seen in this FGRPG1 plot,
...and this GW plot.
Zooming in on the <300 Hz data does show some negative correlation, but I didn't run the numbers. General task times for each frequency range set can be eye-balled on the x-axis. Times span ranges with long tails, so I'm not sure what statistical metric would be best to compare frequency sets.
EDIT: changed <200 Hz to <300 Hz
Ideas are not fixed, nor should they be; we live in model-dependent reality.
I've never seen the need to graph anything out or install anything extra. The tasks per day/per month for each host is available for each project I run through the simple BoincTasks interface.
Agreed. That is primarily what I also look for.
I just made the suggestion to graph different hosts so we could make a comparison visually as opposed to mentally.
I've never seen the need to graph anything out or install anything extra. The tasks per day/per month for each host is available for each project I run through the simple BoincTasks interface.
Aha, the last time I looked, I got the impression that BoincTasks was a Windows-only app. I see now that it comes in Linux flavor. I'll give it a try.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
I've never seen the need to graph anything out or install anything extra. The tasks per day/per month for each host is available for each project I run through the simple BoincTasks interface.
Aha, the last time I looked, I got the impression that BoincTasks was a Windows-only app. I see now that it comes in Linux flavor. I'll give it a try.
It doesn't come in a Linux flavor actually. You just run the Windows app under Wine in Linux.
There is a OS-agnostic Javascript version which was a bit fiddly when I tried the first iteration. Think a lot of the initial bugs got worked out from my feedback.
It doesn't come in a Linux flavor actually. You just run the Windows app under Wine in Linux.
There is a OS-agnostic Javascript version which was a bit fiddly when I tried the first iteration. Think a lot of the initial bugs got worked out from my feedback.
Yes, I tried the various download links at that site, but after fiddling about I eventually discovered that the developer, eFMer, has made it available as a Snap package. So, from a search in the Ubuntu Software app, I got it installed and running out-of-the-box with a couple of clicks.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
cecht wrote: This is just a
)
Cecht, you're doing a great job at making a graph to plot the frequency and tasks/day, and possibly others. And that is all well and good. But, not to rain on your parade, why do this? To what extent is this going to be useful to anyone within the BOINC community, or to be more specific, Einstein@home?
Unless you're getting your toes wet before trying to tackle another BOINC project, or some other tasks like another program, I don't quite see why this particular project is important. I realize that monitoring the project frequency and number of tasks/day may be good for you. I just am having trouble understanding the why, or to what extent.
Proud member of the Old Farts Association
It’s interesting to see the
)
It’s interesting to see the frequency plotted for each series.
but maybe something got misinterpreted when I suggested looking at the frequencies. I wanted to see avg runtime for the frequency ranges. Not just the frequencies themselves. Which means grouping the task results by frequency then plotting the runtimes for each frequency range.
_________________________________________________________________________
GWGeorge007 wrote: To what
)
Fair enough. The original idea of the counting program was use task completion rate as a measure of productivity because completion rate is not easily obtained from BoincManager or a user's account page. I've used it to compare hosts, Projects and GPU settings and thought others might also want to do that. Also, i get a kick knowing how many tasks my hosts can grind through in a day, month, or year.
The plotting program I'm developing is a E@H data exploration tool. At the moment, I'm not sure what usefulness it might have.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
cecht wrote: I've used it to
)
Ahh... I now am beginning to see some of the benefits of it, to an extent.
If you could, can you retrieve the data results from differing hosts? That is to say, it would be useful to compare different hosts as far as productivity, or number of tasks each host grinds through per day/month/year for the same project.
If the plotting program were to be useful (at least to me), I would think if it had an option to graph out the results over time as opposed to plotting dots, and also comparing different hosts, then I could see the benefit.
Proud member of the Old Farts Association
I've never seen the need to
)
I've never seen the need to graph anything out or install anything extra. The tasks per day/per month for each host is available for each project I run through the simple BoincTasks interface.
Ian&Steve C. wrote:It’s
)
I re-jigged the frequency plots to run-time vs. frequency for FGRPG1 (N = 139744) and GW O3 (N = 119607) tasks. In short, except perhaps for limited frequency ranges or data series, there is no correlation. Spearman's correlation for FGRPG1 data is -0.63 and for GW O3 is -0.44.
I think the plotting program is best for data exploration (see screenshots below). If you or anyone can think of some E@H plotting or reporting feature that might be useful, I'll be glad (?) to give it a shot.
The poor or absent correlations are seen in this FGRPG1 plot,
...and this GW plot.
Zooming in on the <300 Hz data does show some negative correlation, but I didn't run the numbers. General task times for each frequency range set can be eye-balled on the x-axis. Times span ranges with long tails, so I'm not sure what statistical metric would be best to compare frequency sets.
EDIT: changed <200 Hz to <300 Hz
Ideas are not fixed, nor should they be; we live in model-dependent reality.
Keith Myers wrote: I've
)
Agreed. That is primarily what I also look for.
I just made the suggestion to graph different hosts so we could make a comparison visually as opposed to mentally.
Proud member of the Old Farts Association
Keith Myers wrote: I've
)
Aha, the last time I looked, I got the impression that BoincTasks was a Windows-only app. I see now that it comes in Linux flavor. I'll give it a try.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
cecht wrote: Keith Myers
)
It doesn't come in a Linux flavor actually. You just run the Windows app under Wine in Linux.
There is a OS-agnostic Javascript version which was a bit fiddly when I tried the first iteration. Think a lot of the initial bugs got worked out from my feedback.
Download BoincTasks.js
Keith Myers wrote: It
)
Yes, I tried the various download links at that site, but after fiddling about I eventually discovered that the developer, eFMer, has made it available as a Snap package. So, from a search in the Ubuntu Software app, I got it installed and running out-of-the-box with a couple of clicks.
Ideas are not fixed, nor should they be; we live in model-dependent reality.