Difference?

[sunu]

I do plain linear regression. mainly to prove that credit new is not linear
0.188 credit/second on beta with some flavour of x37.

I have thousands of data if you want help.

What method do you use for credit/sec?

[Miep]

I do plain linear regression. mainly to prove that credit new is not linear
0.188 credit/second on beta with some flavour of x37.

I have thousands of data if you want help.

What method do you use for credit/sec?

not something that's suitable for large amount of data - unless it is in some form that can be read by excel.

As I said - linear regression. Stuff the data into Excel, make a scatter graph and tell excel to 'add trendline, linear, with equation and r value showing'

Should be fairly simple to code, but my Fortran is rusty and I don't have access to a compiler or NAG libraries anyway. Not to mention my 15 year old scripts on how to manually calculate the smallest sum of squared errors and the precise formula for linear regression...

[sunu]

Computer is http://setiathome.berkeley.edu/show_host_detail.php?hostid=3281360

If a task is made from more than 1 GPUs (boinc restart, preemption) it goes to the GPU that finishes it.

                              Sample
Device0 = GTX295 Core1         1254
Device1 = GTX295 Core2         1242
Device2 = GTX285               1490

[Miep]

hmm. Runtime to AR may be scattering too much and credit to runtime scatters, so I wouldn't expect credit/sec to AR to follow a linear dependency.

Anyway we know that runtime to AR is not linear over the whole AR range, but only over shorter AR stretches. There is a very old plot of runtime/AR with lines across somewhere in NC.

Try plotting credit to runtime and linear regression on that.

[sunu]

Data same as above

[Claggy]

Remember the runtime and credit awarded might be from your wingman's result, and not yours

Claggy

[sunu]

Remember the runtime and credit awarded might be from your wingman's result, and not your's,
Claggy

? ? ? ? ? ? ? ? ? ?

[Claggy]

Remember the runtime and credit awarded might be from your wingman's result, and not your's,
Claggy

? ? ? ? ? ? ? ? ? ?

Aspecially if you're running anonymous platform,

Claggy

[sunu]

What do you mean? That credit is decided with accordance to your wingman? Everyone knows and accepts that.

[Claggy]

What do you mean? That credit is decided with accordance to your wingman? Everyone knows and accepts that.

You'll have to have a read of NewCredit and try and understand it, i don't fully understand it myself, let alone anyone else,
but from watching the amount of Credit awarded on Astropulse tasks, we used to get 700 to 800 credits with the old credit way,
with new credit my anonymous platform hosts get 700 to 800 Credit when matched to hosts app_versions that have more than 10 validations using the Stock app,
and 500 to 600 Credits when matched to hosts that haven't done 10 validations on that app_version using the Stock app,
and 500 to 600 Credits when matched to other hosts that are running anonymous platform,

Claggy

[Miep]

I haven't been bored enough to try and go through the finer maths of it and then go through the code...

anyway, his runtime should be his

dependency on wingmate is deliberate, I just don't believe the statistics used are sound - or the code. Rs of .85  and lower are linear only to a biologist...