User:Carl Boettiger/Notebook/Stochastic Population Dynamics/2010/08/15

{| width="800"
 * style="background-color: #EEE"|[[Image:owwnotebook_icon.png|128px]] Stochastic Population Dynamics
 * style="background-color: #F2F2F2" align="center"|  |Main project page
 * style="background-color: #F2F2F2" align="center"|  |Main project page


 * colspan="2"|
 * colspan="2"|

Warning Signals
Exploring convergence in maximum likelihood model fits each on data produced by their own simulation method:

Error in optim(pars, warning.likfn, method = method, lower = lower) : non-finite finite-difference value [3]
 * fit1, basic OU method, converges just fine. (using L-BFGS-B method with appropriate bounding box to avoid negatives).
 * fit2, the linear change rate model, converges fine using Nelder-Meade (using 1e12 ~ inf likelihood return when attempting negative alpha or sigma), but throws hard errors for L-BFGS-B:


 * fit3 runs on "L-BFGS-B" and returns reasonable values, though not meeting convergence standards. Converges successfully on Nelder-Mead.


 * using this version of warning_example.R for the example exploration and
 * using this version of methods from sde_likelihood.R

Code notes

 * Because optim call takes just parameters and not data, writing data directly to the environment as X so it can be found, following the example of sde package. However this means in functions that want to pass the data to optim calls, they must write the data to the appropriate level, right


 * }