Closed daw538 closed 5 years ago
The epsilon is passing through the epsilon = 0 boundary and then continues from 1.0. change the priors to -0.5 to 1.0.
G
On Thu, 21 Mar 2019, 15:42 daw538, notifications@github.com wrote:
Assigned #23 https://github.com/daw538/y4project/issues/23 to @grd349 https://github.com/grd349.
— You are receiving this because you were assigned.
Reply to this email directly, view it on GitHub https://github.com/daw538/y4project/issues/23#event-2220156961, or mute the thread https://github.com/notifications/unsubscribe-auth/AOGXCc3mGkiL4EihBc8qbvPaN7MuAQjXks5vY6hMgaJpZM4cBvZ1 .
--
Dr Guy R. Davies Lecturer in Astrophysics School of Physics and Astronomy The University of Birmingham Edgbaston Birmingham B15 2TT
Tel +44 (0) 121 414 4597
g.r.davies@bham.ac.uk grd349@gmail.com davies@bison.ph.bham.ac.uk
That's what I suspected, but nothing immediately shouts out at me stopping epsilon going below 0 in our code.
real eps_std[N];
real<lower=0> eps_sig;
real epsA;
real epsB;
epsilon[i] = eps_std[i] * eps_sig + (epsA + epsB * log(dnu[i]));
eps_std ~ normal(0, 1);
eps_sig ~ normal(0, 0.5);
epsA ~ normal(0.601, 0.25);
epsB ~ normal(0.632, 0.25);
Presumably its the <lower=0>
on eps_sig that is restricting it. (I only want to check as I don't want to waste time now compiling and testing the wrong thing!) :smiley:
EDIT: I'm running a few tests at the moment.
Just analysed the parameter plot for NGC6819 and the correlation plots on the whole look as expected with the exception of epsilon (I can deal with the outliers in the other parameters easily). As you will see in this image
Before I do anything to the code, is there anything that would explain the bunch in the top left of the plot? As you can see it pulls the power law completely out of kilter, and I think this is what is causing some of the 'tau models' to deteriorate. I have checked this with more than one run and the results are consistent with each other, so is there some understanding I'm missing or should I simply throw the outliers away?