Adding large GPMSA verification example #615
Conversation
|
|
||
| // Read in data and store the standard deviation of the simulation | ||
| // data (ignored for now). | ||
| readData("sim_scalar.dat", "y_exp_scalar.txt", simulationScenarios, paramVecs, |
There was a problem hiding this comment.
Are these files included in the PR anywhere?
There was a problem hiding this comment.
Wait, never mind, those are the files that were merged in #580.
dmcdougall
force-pushed
the
gpmsa_pdf_verif_large
branch
from
0b19994 to
f50e7fa
Compare
Codecov Report
@@ Coverage Diff @@
## dev #615 +/- ##
==========================================
- Coverage 75.07% 74.87% -0.21%
==========================================
Files 311 312 +1
Lines 23528 23757 +229
==========================================
+ Hits 17664 17788 +124
- Misses 5864 5969 +105
Continue to review full report at Codecov.
|
Also converts some tabs to spaces.
dmcdougall
force-pushed
the
gpmsa_pdf_verif_large
branch
from
326b425 to
a5d4fb7
Compare
For GPMSA, we expect our log prior and matlab's log prior to differ only by an additive constant in log space. I've added a comment that explains this.
|
@roystgnr Ok I think I'm done here too. Let me know if you have any comments. |
|
@briadam This is the pull request where I destroyed all your hard work. I'll resurrect it. |
|
Hardly destroyed... just morphed it into a necessary prerequisite test. ;-) Honestly, you'll probably do better to just write an inverse problem departing from scalar_pdf_large without looking at what I did as you'll likely then immediately see the issues I ran into with the initial positions and proposal covariance. I can see having to scale the proposal covariance as we discussed today, but it's possible at this point (given all the work you've done) that if you take the initial point at the mean and proposal covariance to be the prior variance, you may fare better than I did, and it'll just work. |
|
Well, it'll be a pull request in a couple of minutes and then you can tell me what to do with it :) |
Replaces #607.