add second method to burnin

[tedvh]

• Convert the burnin approach that Nathan had to method 1
• Add a new method (#2) that uses the larger step size only and which uses the older techniques in BASE-8 to decorrelate distance and age. This correlation is the likely cause of sampling problems for WDs.

[argiopetech]

Okay, It's been awhile since we looked at this. Can you confirm that burnin approach 1 is the current approach whereby we do half the burnin with big steps and the second half with smaller steps?

Additionally, 6bf8433a4aa977b5b45d2cac49e64dd151dc6808 contains the "--bigStepBurnin" flag which causes only the large step size to run (like we've been doing with WDs), but does not include any techniques from BASE-8. Is this the desired functionality, or do I need to go consult BASE-8?

[tedvh]

I think those are the first two options. We have reason to believe that option 1 works for open clusters where one is fitting the IFMR, as that is what Nathan was using. Experimentally, we have found that option 2 works OK for individual WDs. I suggest adding option 3, which adjusts the step sizes based on the acceptance ratio, as we did with BASE-8 (mcmc):

/////////////////////////// //// Adjust Step Sizes //// /////////////////////////// double adjustStepSizes(int accept, int reject){

int i, iupper = 4, ilower = 7; double fracAccept; double fa[7] = {0.9, 0.7, 0.5, 0.3, 0.05, 0.15, 0.2}; double w[7] = {2.0, 1.8, 1.5, 1.2, 0.5, 1/1.8, 1/1.5};

if(accept + reject < 200) return 1.0;

fracAccept = _accept / (double) (_accept + reject); accept = 0; *reject = 0;

for(i=0;i<iupper;i++){ if(fracAccept > fa[i]) return w[i]; } for(i = 4; i < ilower; i++){ if(fracAccept < fa[i]) return w[i]; } return 1.0; } // adjustStepSizes

We will probably want other burnin approaches as well, but this is enough for 9.3.0.

-Ted

On 7/25/13 4:32 PM, Elliot Robinson wrote:

Okay, It's been awhile since we looked at this. Can you confirm that burnin approach 1 is the current approach whereby we do half the burnin with big steps and the second half with smaller steps?

Additionally, 6544890 https://github.com/argiopetech/base/commit/6544890dcd987760dcc205b7964a7ba7bfc963a9 contains the "--bigStepBurnin" flag which causes only the large step size to run (like we've been doing with WDs), but does not include any techniques from BASE-8. Is this the desired functionality, or do I need to go consult BASE-8?

— Reply to this email directly or view it on GitHub https://github.com/argiopetech/base/issues/20#issuecomment-21582521.

Ted von Hippel

Department of Physical Sciences Embry-Riddle Aeronautical University 600 S. Clyde Morris Boulevard Daytona Beach, FL 32114-3900 386-226-7751

[argiopetech]

Option 3 seems to be described in Add Adaptive Burnin (Issue #33) and is currently scheduled for 9.3.1. Do we want to pull it back to 9.3.0?

[tedvh]

Ah, OK. That is fine. 9.3.1 is good.

On 7/25/13 4:50 PM, Elliot Robinson wrote:

Option 3 seems to be described in Add Adaptive Burnin (Issue #33 https://github.com/argiopetech/base/issues/33) and is currently scheduled for 9.3.1. Do we want to pull it back to 9.3.0?

— Reply to this email directly or view it on GitHub https://github.com/argiopetech/base/issues/20#issuecomment-21583685.

Ted von Hippel

Department of Physical Sciences Embry-Riddle Aeronautical University 600 S. Clyde Morris Boulevard Daytona Beach, FL 32114-3900 386-226-7751