Final likelihood differs with user input tree

[GoogleCodeExporter]

The final likelihood computed on a fixed user-provided tree (/w branch lengths) 
differs between PhyML and BEAGLE. Specifically, the following two commands 
result in different final likelihoods:

`./src/phyml-beagle -i ./datasets/toy.txt -d nt -q -c 4 -v 0 -t e -m JC69 -f 
'0.25,0.25,0.25,0.25' -o none -b 0 --r_seed 1999 -u datasets/input_tree_toy.txt`
Final likelihood: -61.682296

`./src/phyml -i ./datasets/toy.txt -d nt -q -c 4 -v 0 -t e -m JC69 -f 
'0.25,0.25,0.25,0.25' -o none -b 0 --r_seed 1999 -u datasets/input_tree_toy.txt`
Final likelihood: -49.881937

If I simply remove the "-u datasets/input_tree_toy.txt" parameter, both of the 
above commands give the same final likelihood (i.e. -49.881937 )

Recall that when the "-u" parameter is provided, PhyML constructs the tree with 
`Read_Tree()`, otherwise it estimates an initial topology with 
`Dist_And_BioNJ()`. 

Recall that BEAGLE has been implemented as "deeply" as possible (i.e. it only 
computes partial likelihoods and is supposedly oblivious to "outside" code), so 
its quite strange that something "high-level" affects the likelihoods... it 
shouldn't matter if the tree is read from a file, or estimated, or XML, or sent 
via socket...

Original issue reported on code.google.com by imran.fa...@gmail.com on 25 Jun 2013 at 11:15

Attachments:

• toy.txt
• input_tree_toy.txt

[GoogleCodeExporter]

Fix. Tips were numbered incorrectly to begin with. The input tree merely 
exposed an existing issue.

Original comment by imran.fa...@gmail.com on 1 Jul 2013 at 10:27

[GoogleCodeExporter]

Original comment by imran.fa...@gmail.com on 19 Jul 2013 at 11:40

• Changed state: Fixed