<html tabindex="0">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
<style style="display: none;" id="owaParaStyle" type="text/css">P {margin-top:0;margin-bottom:0;}</style>
</head>
<body aria-label="Message body" fpstyle="1" dir="ltr">
<div style="font-family: Tahoma;font-size: 10pt;color: #000000;margin: 0;">Dear Gavin, thanks, see my answers to your four points below:<br>
<br>
Cheers<br>
<br>
Ruben<br>
<br>
<div style="font-family:Tahoma; font-size:10pt">
<hr>
Ruben H. Roa-Ureta, Ph. D.<br>
AZTI Tecnalia, Marine Research Division<br>
Txatxarramedi Ugartea z/g, Sukarrieta,<br>
Bizkaia, Spain
<hr>
</div>
<br>
<hr tabindex="-1">
<div id="divRplyFwdMsg"><font style="font-size:11pt" color="#000000" face="Calibri, sans-serif"><b>From:</b> Gavin Fay - NOAA Affiliate<br>
<b>Sent:</b> Wednesday, July 30, 2014 10:23 PM<br>
<b>To:</b> Rubén Roa<br>
<b>Subject:</b> ADMB RE problems<br>
</font><br>
</div>
<div></div>
<div>
<div dir="ltr">
<div>
<div>
<div>Hi Ruben,<br>
<br>
</div>
Based on a quick glance at you tpl file, I note the following:<br>
<br>
</div>
1. You have calculations in the procedure section pertaining to estimated parameters that is not in the separable functions (exponentiating the model parameters). In a separable model, all calculations must be done inside the separable functions.<br>
<br>
R: I read that bit in the ADMB-RE manual but in fact in doesn't matter. Whether I put these exponentiations both in the procedure section and the separable functions, or just in the separable function, the result is the same (except that the exponentiated parameters
are not reported). Pedersen et al's tpl also works with those exponentiations in the procedure section and the separable functions.<br>
<br>
</div>
<div>2. You have multiple declarations of the same variable (i.e. the parameter names in your separable functions are the same as those declared in the PARAMETER_SECTION). I do not know whether this is an issue, but common practice is to modify the names of
parameters in the separable functions to force them to be local (e.g. logK_i rather than logK).
<br>
<br>
R: I didn't know that. However, note that my model now works as I posted it just by adding phases to parameters. These are 1 to 5 for Sdy, SdU, p, r. and K, respectively. Also the lower bound for Sdy needs to be lowered to 8.0.<br>
</div>
<div>
<div>
<div>
<div><br>
</div>
<div>3. Have you tried initializing the random effects at non-zero values and running your model using -noinit ?<br>
<br>
R: Nope, I should try that because these random effects are in the order of tens of thousands.<br>
</div>
<div><br>
</div>
<div>4. Does your model run ok with the random_effects_vector replaced as a init_vector and the separable functions replaced with regular functions? (you will probably have to not estimate the variance term). If compiling this at least runs and gives you answers
(albeit wrong ones) that would tell me that the problem is related to your treatment of separability.<br>
</div>
<div><br>
</div>
<div>R: As I wrote above, the random effects models now works as posted, by adding the phases. The estimates and correlation matrix were quite reasonable.<br>
<br>
Hope this helps.<br>
<br>
</div>
<div>Cheers,<br>
</div>
<div>Gavin<br>
<br>
>><br>
</div>
<div><br>
Message: 2<br>
Date: Sat, 26 Jul 2014 07:31:54 +0000<br>
From: Rub?n Roa <<a href="mailto:rroa@azti.es">rroa@azti.es</a>><br>
To: "<a href="mailto:users@admb-project.org">users@admb-project.org</a>" <<a href="mailto:users@admb-project.org">users@admb-project.org</a>><br>
Subject: [ADMB Users] State-space Pella-Tomlison model goes ballistic<br>
Message-ID: <
<div id=":12a" class="">b4a1d3b0950946329f342034462028db@DSREXCH2013.azti.local><br>
Content-Type: text/plain; charset="iso-8859-1"<br>
<br>
Hi ADMB-experts,<br>
<br>
I'm trying to fit a Pella-Tomlinson model to absolute annual biomass MLEs from another, more basic model, and the catch history of a sea urchin fishery from southern Chile. I took the Theta tpl of Pedersen et al. (2011, Ecological Modelling 222:1394) to adapt
it to my case. The catch history is 28 years long (1986 - 2013) whereas the MLEs of absolute biomass cover the latest 18 years.<br>
<br>
The problem is that although the hyper-parameters seem to have been estimated, the random effects estimates are printed as? -1.#QNBe+00 whereas the standard errors and correlations are printed as -1.#INDe+00 or -1.#IND in the cor file.<br>
<br>
The par file does print the random effect estimates but the max. gradient component is very high.<br>
<br>
I seem to face some run time issues but since my experience with ADMB is very limited and this is my first try at a state-space model with ADMB-RE, I cannot rule out some programming issues, poor initial values, or something I haven't even considered.<br>
<br>
Thanks in advance for any advice if you can take a look at my code.<br>
<br>
Ruben<br>
________________________________<br>
Ruben H. Roa-Ureta, Ph. D.<br>
AZTI Tecnalia, Marine Research Division<br>
Txatxarramedi Ugartea z/g, Sukarrieta,<br>
Bizkaia, Spain<br>
________________________________<br>
-------------- next part --------------<br>
A non-text attachment was scrubbed...<br>
Name: statspacreloxbiom1.tpl<br>
Type: application/octet-stream<br>
Size: 1656 bytes<br>
Desc: statspacreloxbiom1.tpl<br>
URL: <<a href="http://lists.admb-project.org/pipermail/users/attachments/20140726/dea3b94c/attachment-0002.obj" target="_blank">http://lists.admb-project.org/pipermail/users/attachments/20140726/dea3b94c/attachment-0002.obj</a>><br>
-------------- next part --------------<br>
A non-text attachment was scrubbed...<br>
Name: statspacreloxbiom1.dat<br>
Type: application/octet-stream<br>
Size: 505 bytes<br>
Desc: statspacreloxbiom1.dat<br>
URL: <<a href="http://lists.admb-project.org/pipermail/users/attachments/20140726/dea3b94c/attachment-0003.obj" target="_blank">http://lists.admb-project.org/pipermail/users/attachments/20140726/dea3b94c/attachment-0003.obj</a>><br>
</div>
<br clear="all">
<br>
-- <br>
<br>
"The contents of this message are mine personally<br>
and do not necessarily reflect any position of NOAA."<br>
<br>
<br>
Gavin Fay, Ph.D.<br>
Postdoctoral Research Associate (current position)<br>
National Marine Fisheries Service, Woods Hole MA.<br>
<br>
Assistant Professor (starting September 2014)<br>
University of Massachusetts Dartmouth<br>
Department of Fisheries Oceanography<br>
School for Marine Science and Technology<br>
<br>
Phone: +1-508-495-2046<br>
Email: <a href="mailto:gavin.fay@noaa.gov" target="_blank">gavin.fay@noaa.gov</a><br>
Web: <a href="http://gavinfay.org" target="_blank">http://gavinfay.org</a> </div>
</div>
</div>
</div>
</div>
</div>
</div>
</body>
</html>