GARD terminates with error

skyungyong commented 4 years ago

Dear @spond,

I have encountered another problem while running GARD. Here is the part of the error messge.

Analysis Version: 0.1

type: nucleotide rv: GDD How many site rate classes to use (permissible range = [2,10], default value = 4, integer): rate-classes: 3 Loaded a nucleotide multiple sequence alignment with 13 sequences, 3804 sites (166 of which are variable) from OG266.cds.msa.fa Minimum size of a partition is set to be 23 sites

Fitting the baseline (single-partition; no breakpoints) model

Log(L) = -6684.35, AIC-c = 13441.41 (36 estimated parameters)

Performing an exhaustive single breakpoint analysis

Done with single breakpoint analysis. Best sinlge break point location: 2627 c-AIC = 13321.0561810831

Performing multi breakpoint analysis using a genetic algorithm

Done with 2 breakpoint analysis. Best break point locations: 1607, 2868 c-AIC = 13276.85624460605 Done with 3 breakpoint analysis. Best break point locations: 441, 1417, 2868 c-AIC = 13218.9037700252 Done with 4 breakpoint analysis. Best break point locations: 441, 1417, 2868, 3178 c-AIC = 13205.59962267284 [GARD] 5 breakpoints [ generation 1, total models 5988, 0% converged]. Min (c-AIC) = 13201.7938 [ delta = 3.81], breakpoints at 219, 441, 1417, 2868, 31 [GARD] 5 breakpoints [ generation 2, total models 6019, 1% converged]. Min (c-AIC) = 13201.7938 [ delta = 3.81], breakpoints at 219, 441, 1417, 2868, 3178 HYPHYMPI terminated. Error: Internal error Internal error in _LikelihoodFunction::ConjugateGradientDescent. The function evaluated at current parameter values [-1006515.591962724] does not match the last recorded LF maximum [-1006446.277244668]

Thank you for your help again!

spond commented 4 years ago

Dear @skyungyong,

Could you send me your input data? These internal errors are rare and difficult to debug:(

Best, Sergei

spond commented 4 years ago

Dear @skyungyong,

I have worked quite a bit on optimizing the implementation of GARD and your dataset completes using the develop branch. The results are attached (view using http://vision.hyphy.org/gard).

Best, Sergei

OG266.txt.GARD.json.zip

Analysis Description

GARD : Genetic Algorithms for Recombination Detection. Implements a heuristic approach to screening alignments of sequences for recombination, by using the CHC genetic algorithm to search for phylogenetic incongruence among different partitions of the data. The number of partitions is determined using a step-up procedure, while the placement of breakpoints is searched for with the GA. The best fitting model (based on c-AIC) is returned; and additional post-hoc tests run to distinguish topological incongruence from rate-variation.

Requirements: A sequence alignment.
Citation: Automated Phylogenetic Detection of Recombination Using a Genetic Algorithm, _Mol Biol Evol 23(10), 1891–1901
Written by: Sergei L Kosakovsky Pond
Contact Information: spond@temple.edu
Analysis Version: 0.1

type: nucleotide rv: None

Loaded a nucleotide multiple sequence alignment with 13 sequences, 3804 sites (166 of which are variable) from /home/sergei/Projects/test/OG266.txt Minimum size of a partition is set to be 23 sites

Fitting the baseline (single-partition; no breakpoints) model

Log(L) = -6797.60, AIC-c = 13657.73 (31 estimated parameters)

Performing an exhaustive single breakpoint analysis

Done with single breakpoint analysis. Best sinlge break point location: 2627 c-AIC = 13467.83455256095

Performing multi breakpoint analysis using a genetic algorithm

Done with 2 breakpoint analysis. Best break point locations: 1417, 2627 c-AIC = 13373.70740365584 Done with 3 breakpoint analysis. Best break point locations: 441, 1417, 2627 c-AIC = 13269.35054436436 Done with 4 breakpoint analysis. Best break point locations: 367, 629, 1417, 2627 c-AIC = 13202.7322411751 Done with 5 breakpoint analysis. Best break point locations: 367, 629, 1417, 2868, 3178 c-AIC = 13166.9380267925 Done with 6 breakpoint analysis. Best break point locations: 367, 629, 1607, 2189, 2487, 3178 c-AIC = 13131.67238237345 Done with 7 breakpoint analysis. Best break point locations: 367, 629, 1417, 1710, 2147, 2487, 3178 c-AIC = 13116.45933886882 Done with 8 breakpoint analysis. Best break point locations: 367, 629, 1417, 1752, 2113, 2374, 2939, 3178 c-AIC = 13110.50278190426 Done with 9 breakpoint analysis. Best break point locations: 367, 441, 629, 1417, 1710, 2113, 2374, 2939, 3178 c-AIC = 13113.23102897245

veg / hyphy