How to increase indicator's ESS value
471 views
Skip to first unread message
Lee
Nov 12, 2012, 2:13:30 AM11/12/12
to beast...@googlegroups.com
Hi all
I'm using discrete phylogeographic analysis and having troubles as indicators & skyline.popSize got low ESS value(<150).
i can improve neither of values even employed 200M generation, sampleFreq= 1000
I have 2 questions:
1) Does low skyline.popSize ESS affects indicators value? since i used discrete anaylsis this parameters's ESS went straight down blew 100.
2) How to improve indicators' ESS value? when using non-reversible rate matrix i got worse indicators.
thanks!
i can improve neither of values even employed 200M generation, sampleFreq= 1000
I have 2 questions:
1) Does low skyline.popSize ESS affects indicators value? since i used discrete anaylsis this parameters's ESS went straight down blew 100.
2) How to improve indicators' ESS value? when using non-reversible rate matrix i got worse indicators.
thanks!
Lee
Nov 14, 2012, 5:10:59 AM11/14/12
to beast...@googlegroups.com
Maybe i can be more specific. HKY model and BSP was used and operator's turing and weight was adjusted according to screen outputs.
I've tired to set skyline.popSize' prior uniform and up-limit was given(35,50,70) and results seems there is no up-limit restriction to the parameter.
as my species was genetic unstructured, i'm wondering if had any impact on parameters like indicator.
xml parameter's prior and operator portions pasted below.
any suggestion?
thanks!
=====================================================
<!-- Define operators -->
<operators id="operators">
<scaleOperator scaleFactor="0.941" weight="0.1">
<parameter idref="kappa"/>
</scaleOperator>
<deltaExchange delta="0.047" weight="0.1">
<parameter idref="frequencies"/>
</deltaExchange>
<scaleOperator scaleFactor="0.341" weight="0.1">
<parameter idref="alpha"/>
</scaleOperator>
<scaleOperator scaleFactor="0.434" weight="3">
<parameter idref="ucld.mean"/>
</scaleOperator>
<scaleOperator scaleFactor="0.176" weight="3">
<parameter idref="ucld.stdev"/>
</scaleOperator>
<subtreeSlide size="0.886" gaussian="true" weight="15">
<treeModel idref="treeModel"/>
</subtreeSlide>
<narrowExchange weight="15">
<treeModel idref="treeModel"/>
</narrowExchange>
<wideExchange weight="3">
<treeModel idref="treeModel"/>
</wideExchange>
<wilsonBalding weight="3">
<treeModel idref="treeModel"/>
</wilsonBalding>
<scaleOperator scaleFactor="0.784" weight="3">
<parameter idref="treeModel.rootHeight"/>
</scaleOperator>
<uniformOperator weight="30">
<parameter idref="treeModel.internalNodeHeights"/>
</uniformOperator>
<scaleOperator scaleFactor="0.287" weight="15">
<parameter idref="skyline.popSize"/>
</scaleOperator>
<deltaExchange delta="1" integer="true" weight="6" autoOptimize="false">
<parameter idref="skyline.groupSize"/>
</deltaExchange>
<upDownOperator scaleFactor="0.986" weight="3">
<up>
<parameter idref="ucld.mean"/>
</up>
<down>
<parameter idref="treeModel.allInternalNodeHeights"/>
</down>
</upDownOperator>
<swapOperator size="1" weight="10" autoOptimize="false">
<parameter idref="branchRates.categories"/>
</swapOperator>
<uniformIntegerOperator weight="10">
<parameter idref="branchRates.categories"/>
</uniformIntegerOperator>
<!-- discrete state model operators -->
<scaleOperator scaleFactor="0.986" weight="10">
<parameter idref="geoSiteModel.mu"/>
</scaleOperator>
<scaleOperator scaleFactor="0.786" weight="30" scaleAllIndependently="true" autoOptimize="true">
<parameter idref="rates"/>
</scaleOperator>
<bitFlipOperator weight="30">
<parameter idref="indicators"/>
</bitFlipOperator>
<bitFlipInSubstitutionModelOperator scaleFactor="0.741" weight="30" autoOptimize="true">
<generalSubstitutionModel idref="location.model"/>
<parameter idref="geoSiteModel.mu"/>
</bitFlipInSubstitutionModelOperator>
</operators>
<!-- Define MCMC -->
<mcmc id="mcmc" chainLength="200000000" autoOptimize="true">
<posterior id="posterior">
<prior id="prior">
<uniformPrior lower="7.0" upper="11.4">
<statistic idref="tmrca(untitled1)"/>
</uniformPrior>
<exponentialPrior mean="8.8" offset="0.0">
<statistic idref="tmrca(untitled1)"/>
</exponentialPrior>
<uniformPrior lower="345.0" upper="350.0">
<parameter idref="kappa"/>
</uniformPrior>
<uniformPrior lower="0.1" upper="0.35">
<parameter idref="frequencies"/>
</uniformPrior>
<uniformPrior lower="0.007" upper="0.015">
<parameter idref="alpha"/>
</uniformPrior>
<exponentialPrior mean="0.3333333333333333" offset="0.0">
<parameter idref="ucld.stdev"/>
</exponentialPrior>
<logNormalPrior mean="-4.853632" stdev="0.4" offset="0.0" meanInRealSpace="false">
<parameter idref="ucld.mean"/>
</logNormalPrior>
<uniformPrior lower="7.0" upper="11.4">
<parameter idref="treeModel.rootHeight"/>
</uniformPrior>
<logNormalPrior mean="2.174752" stdev="0.4" offset="0.0" meanInRealSpace="false">
<parameter idref="treeModel.rootHeight"/>
</logNormalPrior>
<logNormalPrior meanInRealSpace="false" offset="0.0" stdev="0.4" mean="3.401197">
<parameter idref="skyline.popSize"/>
</logNormalPrior>
<generalizedSkyLineLikelihood idref="skyline"/>
<exponentialMarkovLikelihood idref="eml1"/>
<!-- discrete state model operators -->
<cachedPrior>
<multivariateGammaPrior>
<data>
<parameter idref="rates"/>
</data>
<meanParameter>
<parameter value="1.000000125 1.251229267 1.868868832 0.890244237 0.398343061 0.212311682 0.203706108 0.288296461 0.732684113 0.184117155 0.748749954 1.351857775 0.629750373 0.518780853 0.222681603 0.196766875 0.308600538 0.680398777 0.187082110 1.660550918 3.077476499 0.476501036 0.250472243 0.243312092 0.360074400 1.485617579 0.214328046 1.131857534 0.506151819 0.236343091 0.218917970 0.335520395 1.072888585 0.200254377 0.493304204 0.268395667 0.264061337 0.394812553 1.914507128 0.228911935 0.367668103 0.262536623 0.639223141 0.663598881 0.271523621 0.519904213 0.796138846 0.298564061 0.963324684 0.429000873 0.274151738 0.712913012 0.472112404 0.467945635 0.245543695"/>
</meanParameter>
<coefficientOfVariation>
<parameter value="1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 "/>
</coefficientOfVariation>
</multivariateGammaPrior>
<parameter idref="rates"/>
</cachedPrior>
<exponentialPrior mean="1" offset="0">
<parameter idref="geoSiteModel.mu"/>
</exponentialPrior>
<!-- BSSVS truncated Poisson prior only use for BSSVS-->
<poissonPrior mean="0.693" offset="10.0">
<statistic idref="nonZeroRates"/>
</poissonPrior>
<uniformPrior lower="0.0" upper="1.0">
<parameter idref="location.frequencies"/>
</uniformPrior>
<!-- START Discrete Traits Model -->
<generalizedSkyLineLikelihood idref="skyline"/>
<generalSubstitutionModel idref="location.model"/>
</prior>
<likelihood id="likelihood">
<treeLikelihood idref="treeLikelihood"/>
<ancestralTreeLikelihood idref="location.treeLikelihood"/>
</likelihood>
</posterior>
<operators idref="operators"/>
I've tired to set skyline.popSize' prior uniform and up-limit was given(35,50,70) and results seems there is no up-limit restriction to the parameter.
as my species was genetic unstructured, i'm wondering if had any impact on parameters like indicator.
xml parameter's prior and operator portions pasted below.
any suggestion?
thanks!
=====================================================
<!-- Define operators -->
<operators id="operators">
<scaleOperator scaleFactor="0.941" weight="0.1">
<parameter idref="kappa"/>
</scaleOperator>
<deltaExchange delta="0.047" weight="0.1">
<parameter idref="frequencies"/>
</deltaExchange>
<scaleOperator scaleFactor="0.341" weight="0.1">
<parameter idref="alpha"/>
</scaleOperator>
<scaleOperator scaleFactor="0.434" weight="3">
<parameter idref="ucld.mean"/>
</scaleOperator>
<scaleOperator scaleFactor="0.176" weight="3">
<parameter idref="ucld.stdev"/>
</scaleOperator>
<subtreeSlide size="0.886" gaussian="true" weight="15">
<treeModel idref="treeModel"/>
</subtreeSlide>
<narrowExchange weight="15">
<treeModel idref="treeModel"/>
</narrowExchange>
<wideExchange weight="3">
<treeModel idref="treeModel"/>
</wideExchange>
<wilsonBalding weight="3">
<treeModel idref="treeModel"/>
</wilsonBalding>
<scaleOperator scaleFactor="0.784" weight="3">
<parameter idref="treeModel.rootHeight"/>
</scaleOperator>
<uniformOperator weight="30">
<parameter idref="treeModel.internalNodeHeights"/>
</uniformOperator>
<scaleOperator scaleFactor="0.287" weight="15">
<parameter idref="skyline.popSize"/>
</scaleOperator>
<deltaExchange delta="1" integer="true" weight="6" autoOptimize="false">
<parameter idref="skyline.groupSize"/>
</deltaExchange>
<upDownOperator scaleFactor="0.986" weight="3">
<up>
<parameter idref="ucld.mean"/>
</up>
<down>
<parameter idref="treeModel.allInternalNodeHeights"/>
</down>
</upDownOperator>
<swapOperator size="1" weight="10" autoOptimize="false">
<parameter idref="branchRates.categories"/>
</swapOperator>
<uniformIntegerOperator weight="10">
<parameter idref="branchRates.categories"/>
</uniformIntegerOperator>
<!-- discrete state model operators -->
<scaleOperator scaleFactor="0.986" weight="10">
<parameter idref="geoSiteModel.mu"/>
</scaleOperator>
<scaleOperator scaleFactor="0.786" weight="30" scaleAllIndependently="true" autoOptimize="true">
<parameter idref="rates"/>
</scaleOperator>
<bitFlipOperator weight="30">
<parameter idref="indicators"/>
</bitFlipOperator>
<bitFlipInSubstitutionModelOperator scaleFactor="0.741" weight="30" autoOptimize="true">
<generalSubstitutionModel idref="location.model"/>
<parameter idref="geoSiteModel.mu"/>
</bitFlipInSubstitutionModelOperator>
</operators>
<!-- Define MCMC -->
<mcmc id="mcmc" chainLength="200000000" autoOptimize="true">
<posterior id="posterior">
<prior id="prior">
<uniformPrior lower="7.0" upper="11.4">
<statistic idref="tmrca(untitled1)"/>
</uniformPrior>
<exponentialPrior mean="8.8" offset="0.0">
<statistic idref="tmrca(untitled1)"/>
</exponentialPrior>
<uniformPrior lower="345.0" upper="350.0">
<parameter idref="kappa"/>
</uniformPrior>
<uniformPrior lower="0.1" upper="0.35">
<parameter idref="frequencies"/>
</uniformPrior>
<uniformPrior lower="0.007" upper="0.015">
<parameter idref="alpha"/>
</uniformPrior>
<exponentialPrior mean="0.3333333333333333" offset="0.0">
<parameter idref="ucld.stdev"/>
</exponentialPrior>
<logNormalPrior mean="-4.853632" stdev="0.4" offset="0.0" meanInRealSpace="false">
<parameter idref="ucld.mean"/>
</logNormalPrior>
<uniformPrior lower="7.0" upper="11.4">
<parameter idref="treeModel.rootHeight"/>
</uniformPrior>
<logNormalPrior mean="2.174752" stdev="0.4" offset="0.0" meanInRealSpace="false">
<parameter idref="treeModel.rootHeight"/>
</logNormalPrior>
<logNormalPrior meanInRealSpace="false" offset="0.0" stdev="0.4" mean="3.401197">
<parameter idref="skyline.popSize"/>
</logNormalPrior>
<generalizedSkyLineLikelihood idref="skyline"/>
<exponentialMarkovLikelihood idref="eml1"/>
<!-- discrete state model operators -->
<cachedPrior>
<multivariateGammaPrior>
<data>
<parameter idref="rates"/>
</data>
<meanParameter>
<parameter value="1.000000125 1.251229267 1.868868832 0.890244237 0.398343061 0.212311682 0.203706108 0.288296461 0.732684113 0.184117155 0.748749954 1.351857775 0.629750373 0.518780853 0.222681603 0.196766875 0.308600538 0.680398777 0.187082110 1.660550918 3.077476499 0.476501036 0.250472243 0.243312092 0.360074400 1.485617579 0.214328046 1.131857534 0.506151819 0.236343091 0.218917970 0.335520395 1.072888585 0.200254377 0.493304204 0.268395667 0.264061337 0.394812553 1.914507128 0.228911935 0.367668103 0.262536623 0.639223141 0.663598881 0.271523621 0.519904213 0.796138846 0.298564061 0.963324684 0.429000873 0.274151738 0.712913012 0.472112404 0.467945635 0.245543695"/>
</meanParameter>
<coefficientOfVariation>
<parameter value="1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 "/>
</coefficientOfVariation>
</multivariateGammaPrior>
<parameter idref="rates"/>
</cachedPrior>
<exponentialPrior mean="1" offset="0">
<parameter idref="geoSiteModel.mu"/>
</exponentialPrior>
<!-- BSSVS truncated Poisson prior only use for BSSVS-->
<poissonPrior mean="0.693" offset="10.0">
<statistic idref="nonZeroRates"/>
</poissonPrior>
<uniformPrior lower="0.0" upper="1.0">
<parameter idref="location.frequencies"/>
</uniformPrior>
<!-- START Discrete Traits Model -->
<generalizedSkyLineLikelihood idref="skyline"/>
<generalSubstitutionModel idref="location.model"/>
</prior>
<likelihood id="likelihood">
<treeLikelihood idref="treeLikelihood"/>
<ancestralTreeLikelihood idref="location.treeLikelihood"/>
</likelihood>
</posterior>
<operators idref="operators"/>
Reply all
Reply to author
Forward
0 new messages