How to combine models run on different computers?

[BMcLoone]

I am running analyses of a single dataset on two computers. I would like to combine the separate .fpt and .dbf files into one. Is this possible?

Many thanks in advance.

[bacollier]

I don't think you can merge those files in a simple manner like cut and paste, or at least I have never heard of it (but I could be wrong).

[cooch]

I am running analyses of a single dataset on two computers. I would like to combine the separate .fpt and .dbf files into one. Is this possible?

Many thanks in advance.

The Results Browser Window allows you to append model results from prior runs, or if you have created input files by means other than the MARK interface program and have then processed these input files with the MARK numerical program mark.exe. This may be what you're after.

There are 3 files that must be available to append the full model results: an ASCII text output file, an ASCII text residuals file, and a binary variance-covariance file. To append a model, you must select the ASCII output file using the usual Windows file selection procedures. The default file type is ".tmp", but you can find other file names by selecting the "All Files" option. The other 2 files must differ in their names only by a single character, and must have the same file suffix, i.e., the Windows file type must be the same for all 3 files.

Do a search in the MARK help file for 'append'.

[gwhite]

This is not currently possible, but I've had this problem in the past, so will add some code that allows you to do this. The main catch is that I have to be sure that the file being added is exactly the same structure and data, or else the user could really screw up!

Gary

[bacollier]

This is not currently possible, but I've had this problem in the past, so will add some code that allows you to do this. The main catch is that I have to be sure that the file being added is exactly the same structure and data, or else the user could really screw up!

Gary

I stand corrected, looks like it is kind of possible, or soon will be.

Maybe we should all invest in flash drives

[Eurycea]

One solution I use on Windows 7 is Microsoft's free Windows Live Mesh program. I know this is "after the fact" for you, but it helps me keep my MARK files synced in several different places. You need internet access, but you can skip the flash drive step.

apparently there is a new beta program for this:
https://www.mesh.com/welcome/default.aspx

[gwhite]

I have updated the code to let you append one Results File to another. New code is now on the MARK web site. Below is the help file for the Append DBF menu choice.
---------------------------------------------------------------------------------------------------------------------------------------
Append DBF

The Results Browser Window allows you to append all the models in another MARK Results File. However, the data type, number of encounter occasions, number of groups, time intervals, multi-state codes, number of individual covariates, etc., all have to match. Typically, this capability is used when 2 different computers are used to generate models and estimates, and then the 2 Results Files have to be combined.

To combine 2 Results Files, open the master file in the usual way. Then, under the File | Append DBF menu choices, select the Results File to be added to the master (open) Results File. All of the models in the selected file will be added to the master file. If you have duplicates, use the Delete Model option to delete them.

The best way to use this facility is to build the first Results File and run a simple model to verify that everything is working. Then, copy the DBF and FPT files to another computer (or subdirectory for a machine with multiple processors). By copying the Results Files, you more or less guarentee that all of the critical parameters are identical when it is time to append one of these Results File to the master Results File.

There are a myriad of ways to mess up this process, and generate erroneous results. For starters, don't change the model selection method, i.e., both should be either AICc or BIC. As described above, the best approach is to duplicate the original files, and not try to duplicate them by creating them individually.

[cooch]

I have updated the code to let you append one Results File to another. New code is now on the MARK web site...

and also mirrored here:

http://www.phidot.org/software/mark/download/

Use whichever server is fastest...

[BMcLoone]

Great! Thank you very much for making the change and for doing so quickly. I greatly appreciate it.

[steven]

Brian's request arose in the context of our work relating to multi-stratum (MS) models. as noted in my post at the end of december (see sin vs. logit), we were fitting MS models and had no success for many traits, except on the sin scale. since our standard set of models includes parametric models, this link function was problematic. the solution has been to use simulated annealing. since december, we've been engaged in a very(!) computationally intensive (but successful) effort to fit all models with simulated annealing (our standard set of 8 models that we run takes a total of at least 2 weeks of computation when we use simulated annealing). Our analyses of all of the traits are not finished, but the results are very interesting; among other things, I'm now convinced that MS analysis should never be done without simulated annealing, especially in as much as we have discovered that when the same link function is used even non-parametric models with and without simulated annealing can differ in their estimates. of course, such a difference is conceivable, but it is still something different to see the difference arise, especially when the resulting models differ substantially in some of their estimates (we're particularly interested in stratum dynamics, and we have seen some marked differences in psi estimates). I will report more on all of our simulated annealing results when they are done. I have never seen a published comparison of the two approaches in the case of MS (or MR) models, at least a comparison that is as large as the one we are creating. I hope the results will be useful to others

In any case, the computationally-intensive effort to run all analyses with simulated annealing has involved using 4 different computers and being able to run the same DBF and FPT files on different computers and then combine them is an obviously handy thing to be able to do. So, once again, we owe Gary White many thanks and one or more libations of his choice at some future point in time!