Description Usage Arguments Details Value Note Author(s) References See Also Examples
MCMC sampling for DDEPN. Takes an initial network and samples from the posterior. runmcmc
is a wrapper function for multiple calls of mcmc_ddepn
, in case that multiple cores
are used for parallel MCMC runs.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 | mcmc_ddepn(dat, phiorig=NULL, phi=NULL, stimuli=NULL,
th=0.8, multicores=FALSE, outfile=NULL, maxiterations=10000,
usebics=FALSE, cores=2, lambda=NULL, B=NULL,Z=NULL,
samplelambda=NULL, hmmiterations=30, fanin=4,
gam=NULL, it=NULL, K=NULL, burnin=1000, priortype="laplaceinhib",
plotresults=TRUE,always_sample_sf=FALSE, scale_lik=FALSE,
allow.stim.off=TRUE,debug=0,retobj=NULL, implementation="C")
runmcmc(x,dat,phiorig,phi,stimuli,th,multicores,outfile,maxiterations,
usebics,cores,lambda,B,Z,samplelambda,hmmiterations,
fanin,gam,it,K,burnin,priortype,
plotresults=TRUE,always_sample_sf=FALSE,
scale_lik=FALSE, allow.stim.off=TRUE,debug=0,
retobj=NULL, implementation="C")
|
dat |
The data matrix. |
phiorig |
The reference network to compare to. Can be NULL. |
phi |
The start network. Empty if NULL. |
stimuli |
The stimuli list. |
th |
Threshold for inclusion of an edge in the final network. |
multicores |
Use multiple cores. Not used here. |
outfile |
File to which the network should be drawn. |
maxiterations |
Integer. Maximum number of MCMC iterations. |
usebics |
Use bics for model selection. |
cores |
Not used here. |
lambda |
NULL, Numeric or NA. The Prior influence hyperparameter for the laplace prior. If
numeric, used as fixed prior strength or starting value for prior strength sampling
(when |
B |
The Prior information matrix. See |
Z |
Normalisation factor for prior. |
hmmiterations |
Maximum iterations in the HMM. |
fanin |
Integer: maximal indegree for nodes. |
gam |
Prior influence strength in scalefree prior. |
it |
Number of iterations to generate background distribution in scalefree prior. |
K |
Proportionality factor in scalefree prior |
samplelambda |
Numeric or NULL. If NULL, the Laplace hyperparameter |
x |
List containing two items: An adjacency matrix |
burnin |
Integer. Specifies the number of iterations used as burnin phase for
|
priortype |
Character. One of |
plotresults |
Boolean. If TRUE, some statistics are plotted while inhibMCMC is running. |
always_sample_sf |
Boolean. Update scaling factor in inhibMCMC sampling through the whole sampling if TRUE. Keep scaling factor fixed after burn-in if FALSE. |
scale_lik |
Boolean. Perform scaling of the likelihood according to how many data points were used to calculate the overall likelihood. |
allow.stim.off |
Boolean. If TRUE, the stimulus can become passive at some time. This will generate additional reachable system states, in particular all states from the normal state matrix, generated by the propagation, but with the stimulus node set to 0. |
debug |
Numeric. If 0, a status bar indicates the progress of the algorithm. If 1 or 2,
extra information is printed to the console (for |
retobj |
List. The output generated during an inhibMCMC run (see |
implementation |
String. One of |
Usually this function is called internally by ddepn
.
A list of the following elements:
phi |
The inferred network. |
L |
Likelihood. |
aic |
Akaikes Information Criterion. |
bic |
Bayesian Information Criterion. |
posterior |
Posterior probability. |
dat |
The data matrix. |
theta |
The parameter matrix for the gaussians. |
gamma |
The state transition matrix. |
gammaposs |
The theoretical state transition matrix, as generated by the effect propagation. |
tps |
A list. Each element is a vector of time points for each experiment in the data matrix. |
stimuli |
List of stimuli. |
reps |
Number of replicates for each experiment. |
hmmiterations |
Maximum number of iterations during an HMM run. |
lastmove |
Type of the last change that was performed. |
coords |
Position in the network where the last change was performed. |
lambda |
Laplace prior hyperparameter. |
B |
Laplace prior matrix. |
Z |
Laplace prior normalisation factor. (Not used at the moment.) |
pegm |
Probability of performing the last move. |
pegmundo |
Probability of reverting the last move. |
nummoves |
Total number of possible moves in the current step. |
fanin |
Maximal indegree for nodes. |
gam |
Sparsity prior hyperparameter. |
it |
Sparsity prior iterations. |
K |
Sparsity prior scaling factor. |
conf.act |
Matrix of beliefs that an edge is an activation. (equals freqa/eoccur) |
conf.inh |
Matrix of beliefs that an edge is an inhibition. (equals freqi/eoccur) |
eoccur |
Matrix of total occurrences of edges. |
phi.orig |
Adjacency of reference network, if given. |
stats |
Matrix of scores and statistics recorded during MCMC. |
freqa |
Counts how often an edge was an activation. |
freqi |
Counts how often an edge was an activation. |
mu_run |
The running mean values of the parameters in theta. |
Qi |
A helper for the calculation of sd_run. |
sd_run |
The running standard deviations of the parameters in theta. |
TODO
Christian Bender
TODO
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 | ## Not run:
## load package
library(ddepn)
## sample a network
n <- 6
signet <- signalnetwork(n=n, nstim=2, cstim=0, prop.inh=0.2)
phit <- signet$phi
stimuli <- signet$stimuli
## sample data
dataset <- makedata(phit, stimuli, mu.bg=1200, sd.bg=400,
mu.signal.a=2000, sd.signal.a=1000)
## prior normalisation factor
lambda <- 0.01
## network to start with
V <- rownames(dataset$datx)
phistart <- matrix(0, nrow=n, ncol=n, dimnames=list(V,V))
## use original network as prior matrix
## reset all entries for inhibiting edges
## to -1
B <- phit
B[B==2] <- -1
## now the sampling
ret <- mcmc_ddepn(dataset$datx, phiorig=phit, phi=phistart, stimuli=stimuli,
th=0.8, multicores=FALSE, outfile=NULL, maxiterations=300,
usebics=FALSE, cores=1, lambda=lambda, B=B,
hmmiterations=100, fanin=4, burnin=100, priortype="laplaceinhib")
## End(Not run)
|
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.