-ln goes to infinite with 5 ranges

[GoogleCodeExporter]

Hello,
I've been running Lagrange with no problem. I started with the Python version 
and now I'm running the C++ one (released Aug 17, 2010) on a Mac OS 10.6. The 
program seems to work fine with an unconstrained data set, but whenever I try 
to constrain the areas of connectivity with the rate matrix the -ln goes to 
infinite and I don't get a reconstruction. I tried running the example from the 
manual (4 ranges, 2 periods) and it works fine, but the one I want to run has 5 
ranges and 6 periods. 

Is there any limitation on the number of possible ranges to reconstruct? I know 
7 is the limit from a computational point of view.

I'm attaching my rate matrix and configuration file and will appreciate any 
input. 

Thanks a lot

Simon

Original issue reported on code.google.com by simonuri...@gmail.com on 31 Aug 2010 at 2:55

Attachments:

• configuration_splits_expert_model.txt
• model.rm
• Ranges_expert_C++.txt
• Combined_1000.tre

[GoogleCodeExporter]

I forgot the tree file

Original comment by simonuri...@gmail.com on 31 Aug 2010 at 2:58

Attachments:

• Combined_1000.tre

[GoogleCodeExporter]

Seems like though that your first rate matrix configuration is the problem

1   0   0   1   0
0   1   0   1   0
0   0   1   0   1
1   1   0   1   0
0   0   1   0   1

if i add a number other than zero to even one of the zeros it gets out of 
infinity. so there is something that is constrained there that eliminates a 
reasonable answer. does that make sense?

so 

1   0   0   1   0.0001
0   1   0   1   0
0   0   1   0   1
1   1   0   1   0
0   0   1   0   1

will give a non-inf answer as will 

1   0   0   1   0
0   1   0   1   0.0001
0   0   1   0   1
1   1   0   1   0
0   0   1   0   1

so i would make sure that those constraints are what you want (you can always 
of course constrain certain distributions which might make it easier as well).

Original comment by blackrim on 1 Sep 2010 at 3:10

• Changed state: Done