I've released the v0.04 source code here (take your pick):

http://milkyway.cs.rpi.edu/milkyway/download/mw3_v0.04.zip

http://milkyway.cs.rpi.edu/milkyway/download/mw3_v0.04.tar


The new code contains a new method for modeling the background of the milky way's halo, which we hope will more accurately represent the background. Right now, there is no good model for this -- and finding a good model is one of the goals of this project, so if we determine a good one this will be a major result. We're very interested to see how this new model performs.



There have also been a couple changes to one of the input parameter files, and the output of the application. It will now output individual likelihoods for the background and each stream being modeled, as well as the total likelihood. This will allow us to optimize over them individually and hopefully get better fits for our models.



This source code also includes Kahan summation, which should make the results for the GPUs and CPUs closer.



There are new test parameter files included with the source, and if you want more information about the parameter files and expected results visit the forum thread in the news section.

---

Here's a new example parameter file (changes are bold):


parameters_version: 0.04
number_parameters: 4
background_weight: 0.0
background_parameters[4]: 1.0, 0.571713, 12.312119, 1.0
background_step[4]: 0.02, 0.000004, 0.00008, 0.02
background_min[4]: 0.0, 0.3, 1.0, 0.1
background_max[4]: 3.0, 1.0, 30.0, 3.0
optimize_parameter[4]: 0, 1, 1, 0
number_streams: 3, 5
stream_weight: -3.305187
stream_weight_step: 0.000001
stream_weight_min: -20.0
stream_weight_max: 20.0
optimize_weight: 1
stream_parameters[5]: 148.010257, 22.453902, 0.42035, -0.468858, 0.760579
stream_step[5]: 0.00003, 0.00004, 0.00006, 0.00004, 0.000004
stream_min[5]: 150.0, 2.30, -6.283185307179586, -6.283185307179586, 0.10
stream_max[5]: 229.0, 57.5, 6.283185307179586, 6.283185307179586, 20.0
optimize_parameter[5]: 1, 1, 1, 1, 1
stream_weight: -1.361644
stream_weight_step: 0.000001
stream_weight_min: -20.0
stream_weight_max: 20.0
optimize_weight: 1
stream_parameters[5]: 177.884238, 23.882892, 1.210639, -1.611974, 8.534378
stream_step[5]: 0.00003, 0.00004, 0.00006, 0.00004, 0.000004
stream_min[5]: 150.0, 2.40, -6.283185307179586, -6.283185307179586, 0.10
stream_max[5]: 229.0, 57.5, 6.283185307179586, 6.283185307179586, 20.0
optimize_parameter[5]: 1, 1, 1, 1, 1
stream_weight: -1.361644
stream_weight_step: 0.000001
stream_weight_min: -20.0
stream_weight_max: 20.0
optimize_weight: 1
stream_parameters[5]: 177.884238, 10.882892, 1.210639, -1.611974, 8.534378
stream_step[5]: 0.00003, 0.00004, 0.00006, 0.00004, 0.000004
stream_min[5]: 150.0, 2.40, -6.283185307179586, -6.283185307179586, 0.10
stream_max[5]: 229.0, 57.5, 6.283185307179586, 6.283185307179586, 20.0
optimize_parameter[5]: 1, 1, 1, 1, 1
convolve: 120
sgr_coordinates: 0
aux_bg_profile: 1
wedge: 11
r[min,max,steps]: 16.0, 23.0, 1400
mu[min,max,steps]: 150, 229, 1600
nu[min,max,steps]: -1.25, 1.25, 640
number_cuts: 0


Having the version number at the top will make it easier to modify the parameters file as we test out more models, and make sure we're using the right version of the parameters file. The only other addition is the aux_bg_profile down near the bottom (which will be either 0 for the old model, or 1 for the new model).

---

There's also test_small_aux.sh and test_large_aux.sh which test the new background model. Here's the expected output for the small sized workunits (large sized workunits will be filled in as I crunch them -- they take a few hours on my laptop so it'll be a couple days until they're all done). Also note that there's the background_only_likelihood and stream_only_likelihood -- these are required, as we'll be using them to try different model optimization strategies. And again, values must be at least accurate to 10e-11, and these results need to be written to standard error.


stripe 11 (aux == 1, small):
44090.61418764234258560464
34.16259801227062098405 667.25896879727622490464 336.16783412736663194664
-3.79069318169778135541
-84.31096247061454107552 -4.18580676656694095072 -4.11982130330399343165
-3.34561414990350636245
stock_osx_x86_64: 0.04 double

stripe 12 (aux == 1, small):
55810.90404403187130810693
93.28291463972850294795 693.31476977139550399443 379.87099590293541950814
-3.89350665070662316225
-83.37050111102392691009 -4.28050569961213867032 -4.20893197255954287783
-3.43988114444296932959
stock_osx_x86_64: 0.04 double

stripe 20 (aux == 1, small):
33728.79975938780262367800
10.71234415081209334630 513.40698152694301370502
-3.59861944338664763166
-107.40741411252416526168 -4.13706816784206310444
-3.41883666179971834609
stock_osx_x86_64: 0.04 double

stripe 21 (aux == 1, small):
22388.94457863535717478953
9.54103668867152165944 458.95295766568636963711
-3.41615563366402552603
-42.61801526167820242108 -3.83839703933581910533
-3.22458789359460729074
stock_osx_x86_64: 0.04 double

stripe 79 (aux == 1, small):
37695.67323256157396826893
98.08330586161720532345
-3.48113010724505222981
-16.33247867611922643505
-3.44040321199726717438
stock_osx_x86_64: 0.04 double

stripe 82 (aux == 1, small):
39131.69896788540063425899
101.09116755806067544654
-3.51180016629466429379
-9.51944353854558400485
-3.45721777592385226541
stock_osx_x86_64: 0.04 double

stripe 86 (aux == 1, small):
39239.40136974808410741389
374.13823192303391351743
-3.52147951248316504191
-5.94382123800844830441
-3.44148079719333344073
stock_osx_x86_64: 0.04 double

---

here are the expected results for aux=1 for the large sized workunits.

stripe 11 (aux == 1, large):

stripe 12 (aux == 1, large):

stripe 20 (aux == 1, large):

stripe 21 (aux == 1, large):

stripe 79 (aux == 1, large):

stripe 82 (aux == 1, large):

stripe 86 (aux == 1, large):

(to come).

---

here are the expected results for the small sized workunits and aux == 0.

stripe 11 (aux == 0, small):
0.00034440467776670103
34.16259801227062098405 667.25896879727622490464 336.16783412736663194664
-3.30823499270256338178
-84.31096247061454107552 -4.18580676656694095072 -4.11982130330399343165
-3.12509788380322994783
stock_osx_x86_64: 0.04 double

stripe 12 (aux == 0, small):
0.00044499685869062346
93.28291463972850294795 693.31476977139550399443 379.87099590293541950814
-3.41586621576637483599
-83.37050111102392691009 -4.28050569961213867032 -4.20893197255954287783
-3.22083007667620435654
stock_osx_x86_64: 0.04 double

stripe 20 (aux == 0, small):
0.00047193922911510617
10.71234415081209334630 513.40698152694301370502
-3.12237267987161093430
-107.40741411252416526168 -4.13706816784206310444
-2.98536129276144279743
stock_osx_x86_64: 0.04 double

stripe 21 (aux == 0, small):
0.00022403587279673257
9.54103668867152165944 458.95295766568636963711
-3.01801580202351704685
-42.61801526167820242108 -3.83839703933581910533
-2.88940405712715753239
stock_osx_x86_64: 0.04 double

stripe 79 (aux == 0, small);
0.00007941689465694971
98.08330586161720532345
-3.00955241399873729691
-16.33247867611922643505
-2.94673379590026485175
stock_osx_x86_64: 0.04 double

stripe 82 (aux == 0, small):
0.00017081140733525433
101.09116755806067544654
-3.08191187667341193190
-9.51944353854558400485
-2.98561763573615523981
stock_osx_x86_64: 0.04 double

stripe 86 (aux == 0, small):
0.00064937938973831709
374.13823192303391351743
-3.10468084478402506221
-5.94382123800844830441
-3.02797378771823533228
stock_osx_x86_64: 0.04 double

---

here are the expected results for aux == 0 and the large sized workunits:

stripe 11 (aux == 0, large):

stripe 12 (aux == 0, large):

stripe 20 (aux == 0, large):

stripe 21 (aux == 0, large):

stripe 79 (aux == 0, large):

stripe 82 (aux == 0, large):

stripe 86 (aux == 0, large):

(to come).

---

For how long this current test schedule will continue? I have missed to increase the fitness values.

Uhm, i'm not the best with pc's... but do we have to place this somewhere in our pc so that milkyway@home works better?

Uhm, i'm not the best with pc's... but do we have to place this somewhere in our pc so that milkyway@home works better?

No ... :)

If you want to compile a version for yourself, which is not likely, you would need this package. If you only want to run MW, no need for it at all ...

I run MW on my systems and have run the later versions as I have been able to get the work (or have been issued the work), but don't have this source as I have zero interest in trying to make my own private version ...

Gispel and others like him will D/L the code and work with Travis and Co. to make the code faster and tighter and we, the unwashed masses will get the benefit ... :)

Started looking at the code, and the more I looked at it, the more I decided to send-in some code mods (call them suggestions).

I've just sent several mods to Travis - hope the email address is correct... :-)

---

http://www.joescat.com/boinc/

I didn´t read any about an ATI-Code. Is there something in development by Cluster Physik, who is a little bit quiet the last days. ;-)