uyedaj / bayou

This is the development version of the R package bayou. bayou fits Bayesian reversible-jump multi-optima OU models to phylogenetic comparative data.

Please report any bugs or issues, as this package is currently under development.

Install using the devtools package.

install.packages("devtools")
require(devtools)
install_github("uyedaj/bayou")
require(bayou)

A tutorial is available at:

https://github.com/uyedaj/bayou/blob/master/tutorial.md

The manuscript describing the method can be found at:

https://doi.org/10.1093/sysbio/syu057

bayou's Issues

ERROR IN BAYOU.MCMC: error: matrix multiplication: incompatible matrix dimensions: 74x15 and 1x1 Error in C_weightmatrix(cache, pars) : matrix multiplication: incompatible matrix dimensions: 74x15 and 1x1

Hello,

Thanks in advance for any help and thank you for your awesome tutorial on using this package.

I was using an older version of bayou (1.1.0) earlier today and I got it to run smoothly with no problems. I then realized I was missing a lot of the functions in the documentation so I updated to the most recent version on GitHub (v 2.1). Now I am unable to get the bayou.mcmc function to work like I did earlier.

Using my data and following along with the tutorial you provided, I keep getting this error message when running bayou.mcmc:

fit1 <- bayou.mcmc(bulb_size_tree, bulb_size.vector, SE = SE, model = "OU", prior, new.dir=getwd(), ngen = 10000)
error: matrix multiplication: incompatible matrix dimensions: 74x15 and 1x1
Error in C_weightmatrix(cache, pars) :
matrix multiplication: incompatible matrix dimensions: 74x15 and 1x1

The prior and SE are the same as in the tutorial.

Any suggestions would be appreciated. Thanks for your time.

Thanks
Cody Coyotee

Convergent evolution

Hi,

Is there a way to fit a model where the same theta has multiple origins?

As far as I know It is posible to have two different theta (two different parameters) that have approximately the same value.
But I would like to know where or not it is possible that the same theta (one parameter) with multiple origins in the phylogeny.

Thanks!
Marcial

Error is thrown with zero length branches

If zero length branches are used then the following error gets thrown when executing a slide move on the zero length branch:

Error in if (m.i - 1 == 1) { : argument is of length zero

Error in if (runif(1) < exp(nll - oll + pr2 - pr1 + hr)) { :
missing value where TRUE/FALSE needed

Error with plot functions bayou package

Hi,

I have the following error when using plotSimmap.mcmc(), plotBranchHeatMap() and phenogram.density(). This is my code:

plotSimmap.mcmc(chainOU, burnin = 0.3, lwd = 2, edge.type = "theta", pal = rainbow, pp.cutoff = 0.3)
plotBranchHeatMap(formTree, chainOU, "theta", burnin = 0.3, pal = cm.colors)
phenogram.density(formTree, p50, burnin = 0.3, chainOU, pp.cutoff = 0.3)

And the error:

Error in plotSimmap(x, ...) : unused arguments (edge.color = .colorRamp(allbranches, pal, 100), use.edge.length = F)

I am not sure where is the cause, could you be so kind to help me?

Thank you in advance,

Pablo

Feature request: function to extract the raw value of a trait from each regime

You mentioned that there's an internal function to do that, but it will spit out a bunch of stuff. Is there an easy way to extract just the trait value for the species that belong to a given regime?

@jtonini

Thanks a lot again!

bayou2OUwie error

Hello,

I have successfully run bayou and now would like to export that information into an acceptable OUwie format. I am using the bayou2OUwie function. I have pars directed to the saved bayou run so that it can locate sb, loc and t2. I have the tree I used for the bayou run, and the tip data are in a named vector sorted in the same order at the tree labels.

I receive the following error:

> bayou2OUwie(chain.fit, fit.tree, fit.sorted2)
Error in table(pars$sb) : all arguments must have the same length

I have checked and all lists have 150001 items.

> length(chain.fit$sb)
[1] 150001
> length(chain.fit$loc)
[1] 150001
> length(chain.fit$t2)
[1] 150001

The number of tips in the tree match the number of rows for the tip data.

I cannot figure out why I am getting this error.

I am using bayou v2.1.1, R version 3.5.2.

Any advice appreciated. Thanks for this package!

Cody Coyotee

Prob of shift proportional to branch length maybe broke on dev version?

Runs that previously worked on the CRAN version of bayou are no longer working for me on the dev version. These runs use the probability of a shift proportional branch length prior outlined in some of the documentation. The runs work for a few generations, but ultimately crash. For instance,

prior2 <- make.prior(prunedMBC, dists=list(dalpha="dhalfcauchy", dsig2="dhalfcauchy",dsb="dsb", dk="cdpois", dtheta="dnorm"), param=list(dalpha=list(scale=1), dsig2=list(scale=1), dk=list(lambda=15, kmax=200), dsb=list(bmax=Inf,prob=prunedMBC$edge.length), dtheta=list(mean=mean(avMBC), sd=2)))

mcmc2 <- bayou.makeMCMC(prunedMBC, avMBC, SE=cvMBC, model="OU", prior=prior2, new.dir=TRUE, samp=10^2, chunk=10^3, ticker.freq=10^4, plot.freq=NULL)

system.time(mcmc2$run(ngen=10^7))

returns

Error in sample.int(x, size, replace, prob) : NA in probability vector
Calls: system.time ... .proposalFn -> .moveFn -> .sample -> sample -> sample.int
Timing stopped at: 2099.404 15.264 2117.245

Perhaps something needs to get fixed with this model on the dev version? Or, perhaps I've misspecified something in the code (entirely possible).

plotSimmap.mcmc/phenogram.density when no shifts are large

throws the 1:o too long a vector error. Just need to add an if statement...

summary.bayouMCMC has errors summarizing location

Values >1.

Can not install on mac

I tried to install it using devtools but was not able to get it.
Loading required package: bayou
Warning message:
In library(package, lib.loc = lib.loc, character.only = TRUE, logical.return = TRUE, :
there is no package called ‘bayou’

makeBayouModel should check if the predictor has NAs and impute=FALSE

Currently doesn't give an informative error message.

Numerical errors when the data have a very small variance

bayou.lik can return a infinite likelihood when the distribution of trait values is very small.

Loading a saved BAYOU run

Hello,

I have to run a large number of runs to reach sufficient ES values. However, when I try to save the data either as an Rdata file or with 'saveRDS = T', I cannot reload the data for later analysis. It would be nice to be able to run these analyses and analyze them at a later time. Perhaps this is already implemented but I cannot figure out how if it is possible.

The errors I receive.
`

test <- load.bayou(file = "BAYOU/bayou.chain.rds")
Error in load.bayou(file = "BAYOU/bayou.chain.rds") :
argument "bayouFit" is missing, with no default

test2 <- load("BAYOU/bayou.chain.rds")
Error: bad restore file magic number (file may be corrupted) -- no data loaded
In addition: Warning message:
file ‘bayou.chain.rds’ has magic number 'X'
Use of save versions prior to 2 is deprecated
`

Thanks for any advice on this.

Thanks,
Cody

Likelihood is wrong when alpha is fixed to 0 using 3 point algorithm

Shouldn't have to specify starting parameters when the model has fixed parameters

When the prior has been specified to have "fixed" parameters, instead of simulating variable parameters from the prior, bayou.mcmc requires that the starting parameters be specified, otherwise it will auto select the fixed parameters.

Tutorial does not reflect updates

Currently, it looks like the tutorial does not function, because the package was updated to use a new 'bayou.makeMCMC'.

From what I can tell, this function does not actually start the MCMC sampler, nor is there any obvious documentation on how to do so.

Closing unused connections

If a bayou run is interrupted, these warnings show up. I need to make a way so that runs that are interrupted still produce a fit object that can be loaded, even if the run didn't finish, and the file connections get closed (like diversitree does)

After read the bayou/tutorial.md, i still don't know how to set the parameters?

I hope I can follow the pipeline of the Current Biology paper "Tempo and Pattern of Avian Brain Size Evolution" to analyze my data, but after reading the tutorial.md and methods in the paper, I still don't know how to set the parameters. The methodｓ in paper are as follows,
"We implemented this approach by combining 10 parallel chains of 2 million iterations each with a burn-in proportion of 0.3. We allowed only one shift per branch and the total number of shifts was constrained by means of a conditional Poisson prior with a mean equal to 2.5% of the total number of branches in the tree and a maximum number of shifts equal to 5%. Starting points for MCMC chains were set by randomly drawing a number of shifts from the prior distribution and assigning these shifts to branches randomly drawn from the phylogeny with a probability proportional to the size of the clade descended from that branch. The MCMC was initialized without any birth-death proposals for the first 10,000 generations to improve the fit of the model."

I don't know how to run "combining 10 parallel chains of 2 million iterations with a burin-in proportion of 0.3", and don't know how to set the parameters "We allowed only one shift per branch", also don't how to run "The MCMC was initialized without any birth-death proposals for the first 10,000 generations to improve the fit of the model." Could you me?

Feature request: Function to extract the posterior of the regression line from shiftSummary

Hi @uyedaj ,
just to remind you to see if there's a simple way to extract the posterior distribution of the regression line from the output object to calculate the 95% CI.

@jtonini

make.refFn error for bounded distributions on continuous parameters

Offending line:
test.dists <- suppressWarnings(apply(is.finite(sapply(dists, function(x) c(x(-0.5), x(1.5), x(0.5), x(1)))), 2, function(x) paste(as.numeric(x), collapse = "")))

Quick fix: parallel for steppingstone

I could pull, fix, merge, etc., but it's two lines of code:
Add
if(parallel){
doParallel::registerDoParallel()
}

to the start of the steppingstone function. Should re-enable parallelization, which otherwise doesn't work on my machine.

Error in mcmcOU$run:Error in summary.connection(x) : invalid connection

Hi!
I am using bayou for my dataset, and thanks for providing such a nice tutorial. Till the following step everything is running well.

mcmcOU <- bayou.makeMCMC(WGDTree, WGDdata, SE=0, prior=priorOU,
new.dir="modelOU/", outname="modelOU_r001", plot.freq=NULL)

However, I am getting error when I am trying to run mcmcOU$run(10000). It is throwing an error like the follows.

Error in summary.connection(x) : invalid connection
In addition: Warning messages:
1: In .Internal(gc(verbose, reset, full)) :
closing unused connection 7 (modelOU//modelOU_r001.rjpars)
2: In .Internal(gc(verbose, reset, full)) :
closing unused connection 6 (modelOU//modelOU_r001.pars)
3: In .Internal(gc(verbose, reset, full)) :
closing unused connection 5 (modelOU//modelOU_r001.t2)
4: In .Internal(gc(verbose, reset, full)) :
closing unused connection 4 (modelOU//modelOU_r001.loc)
5: In .Internal(gc(verbose, reset, full)) :
closing unused connection 3 (modelOU//modelOU_r001.sb)

Any help in this regard is highly appreciated. Thanks in advance for your help and time.

Regards,
Tina