search

ddarriba / jmodeltest2

Automatically exported from code.google.com/p/jmodeltest2
GNU General Public License v3.0
74 stars 47 forks source link

negative information criteria values #16

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago 
What steps will reproduce the problem?
1. load the data file
2. specify 11 patterns, exclude +I, increase +G to 8
3. specify AICc, BIC, and DT, after likelihood scores calculated

What is the expected output? What do you see instead?

Program freezes on local machine. Running online, AICc has negative values. BIC 
and DT not computed. K differs across models

What version of the product are you using? On what operating system?

OSX 10.6.8, jmodeltest 2.1.1.

Please provide any additional information below.

Here is the example output:

---------------------------------------------------------------
*                                                             *
*        CORRECTED AKAIKE INFORMATION CRITERION (AICc)        *
*                                                             *
---------------------------------------------------------------

 Model selected: 
   Model = TVM+G
   partition = 012314
   -lnL = 2974.5260
   K = 375
   freqA = 0.1869 
   freqC = 0.3538 
   freqG = 0.2607 
   freqT = 0.1987 
   R(a) [AC] =  1.1003
   R(b) [AG] =  3.8034
   R(c) [AT] =  0.9051
   R(d) [CG] =  0.9751
   R(e) [CT] =  3.8034
   R(f) [GT] =  1.0000
   gamma shape = 0.1270 

* AICc MODEL SELECTION : Selection uncertainty

Model             -lnL    K        AICc      delta      weight cumWeight
------------------------------------------------------------------------ 
TVM+G        2974.5260   375  -275300.9481     0.0000      1.0000    1.0000 
GTR          3517.3863   375  -274215.2274  1085.7207   0.00e+000    1.0000 
GTR+G        2960.7373   376  -135078.5255  140222.4226   0.00e+000    1.0000 
JC           3655.7368   367  46632.9021  321933.8501   0.00e+000    1.0000 
JC+G         3077.3968   368  52154.7937  327455.7417   0.00e+000    1.0000 
K80          3578.3957   368  53156.7913  328457.7394   0.00e+000    1.0000 
K80+G        2994.5908   369  61339.1817  336640.1297   0.00e+000    1.0000 
TrNef        3536.0330   369  62422.0661  337723.0141   0.00e+000    1.0000 
TPM3         3575.8030   369  62501.6061  337802.5541   0.00e+000    1.0000 
TPM2         3577.6094   369  62505.2188  337806.1668   0.00e+000    1.0000 
TPM1         3578.3616   369  62506.7232  337807.6713   0.00e+000    1.0000 
TrNef+G      2980.1488   370  75335.2975  350636.2456   0.00e+000    1.0000 
TPM3+G       2992.1638   370  75359.3276  350660.2757   0.00e+000    1.0000 
TPM2+G       2993.5295   370  75362.0591  350663.0071   0.00e+000    1.0000 
TPM1+G       2994.5600   370  75364.1199  350665.0680   0.00e+000    1.0000 
TIM3ef       3533.3276   370  76441.6551  351742.6032   0.00e+000    1.0000 
TIM2ef       3535.2259   370  76445.4518  351746.3998   0.00e+000    1.0000 
TIM1ef       3536.0039   370  76447.0077  351747.9558   0.00e+000    1.0000 
F81          3645.2379   370  76665.4758  351966.4238   0.00e+000    1.0000 
TIM3ef+G     2977.3015   371  98704.6031  374005.5511   0.00e+000    1.0000 
TIM2ef+G     2978.7527   371  98707.5053  374008.4534   0.00e+000    1.0000 
TIM1ef+G     2980.1399   371  98710.2797  374011.2278   0.00e+000    1.0000 
F81+G        3054.3464   371  98858.6928  374159.6409   0.00e+000    1.0000 
HKY          3561.0965   371  99872.1930  375173.1411   0.00e+000    1.0000 
TVMef        3574.9119   371  99899.8238  375200.7718   0.00e+000    1.0000 
HKY+G        2974.7730   372  145449.5459  420750.4940   0.00e+000    1.0000 
TVMef+G      2990.9115   372  145481.8230  420782.7711   0.00e+000    1.0000 
TrN          3517.7249   372  146535.4498  421836.3978   0.00e+000    1.0000 
SYM          3532.3952   372  146564.7903  421865.7384   0.00e+000    1.0000 
TPM3uf       3560.7858   372  146621.5715  421922.5196   0.00e+000    1.0000 
TPM2uf       3561.0500   372  146622.1001  421923.0482   0.00e+000    1.0000 
TPM1uf       3561.0688   372  146622.1376  421923.0856   0.00e+000    1.0000 
TrN+G        2961.0899   373  285672.1798  560973.1278   0.00e+000    1.0000 
TPM1uf+G     2974.6176   373  285699.2352  561000.1832   0.00e+000    1.0000 
TPM3uf+G     2974.7166   373  285699.4332  561000.3813   0.00e+000    1.0000 
TPM2uf+G     2974.7471   373  285699.4942  561000.4422   0.00e+000    1.0000 
SYM+G        2975.6656   373  285701.3313  561002.2793   0.00e+000    1.0000 
TIM2         3517.5914   373  286785.1828  562086.1309   0.00e+000    1.0000 
TIM3         3517.6407   373  286785.2814  562086.2294   0.00e+000    1.0000 
TIM1         3517.6916   373  286785.3832  562086.3313   0.00e+000    1.0000 
TIM1+G       2961.0297   374

Original issue reported on code.google.com by jssden...@gmail.com on 1 Mar 2013 at 9:34

Attachments:

GoogleCodeExporter commented 9 years ago 
Thanks for the report. I'll check this as soon as possible. Where did you run 
it "online"?

Original comment by DiegoDL84 on 2 Mar 2013 at 11:06

GoogleCodeExporter commented 9 years ago 
Thanks for the update. I've been trying to relocate the site that ran it as
a GUI, but haven't been able to so far. It was a link in a discussion group
about speed.

Original comment by jssden...@gmail.com on 2 Mar 2013 at 5:29

GoogleCodeExporter commented 9 years ago 
I could execute it locally without problems, and I found what was happening. 

Actually there is no error in AICc, but sample size for your alignment is too 
small compared to the number of parameters of the models, since there is a lot 
of taxa compared to the length of each sequence. Default sample size is the 
alignment length (375), and the number of parameters of GTR model is 375 too: 5 
rate parameters + 3 frequencies + (185*2-3) branch lengths. Thus, the part of 
the AICc formula which says (sample size - parameters - 1) produces a negative 
result.

I encourage you to use a higher sample size like NxL or Shannon entropy.

Original comment by DiegoDL84 on 4 Mar 2013 at 2:50