[ADMB Users] lgam in 10.0
davef at otter-rsch.com
Sun May 15 11:42:40 PDT 2011
It raining again (still).
Thought I would look at this.
Below is the whole TPL file as I have modified it to get it to run.
first problme is that posfun must have a non negative second argument.
This is where the "positiveness" kicks in. If you make it 0.0 the
curvature is infinite. Best is to tune it as the model gets fitted.
Call it eps2 and begin with a bigger number in the first phase.
This is a typical technique to make the objective function well
behaved until you get near the solution.
will cause you trouble if p is too close to 0 or 1. so add a tiny number
and tune it as well.
Note the order of addition (eps1+(1-p)) ans figure out why (eps1+1-p)
may not work.
Then add a kludge variable with the number of phases that you want. I
always name if
kkludge so that I l know what its purpose is.
Now run the model and tune the values of eps1 eps2 to be small but the
model still converges.
I got into this and ended up with 5 phases. the value of eps1 has a
I ran with the CLO of -crit 1.e-8.
else if (current_phase()<3)
else if (current_phase()<4)
else if (current_phase()<5)
arrmblsize = 2147483647;
More information about the Users