rosalia: Fit a Markov network to binary data
In davharris/rosalia: Exact inference for small binary Markov networks

Description Usage Arguments Value Examples

The optimization is performed using the BFGS method in the optim function. Note that the likelihood function includes 2^N terms for N nodes, so parameter estimation can be very slow with more than about 20-25 nodes.

1 2	rosalia(x, prior = make_logistic_prior(location = 0, scale = 1), maxit = 100, trace = 1, hessian = FALSE, parlist, ...)

`x`	A binary matrix. Each row corresponds to an observed sample and each column corresponds to one node in the observed network.
`prior`	An object of class `prior`. The default prior, produced by `make_logistic_prior`, regularizes the parameter estimates. This is important in Markov networks because the number of parameters will often be large compared to the number of independent observations. The flat prior produced by `make_logistic_prior` is unbiased, but can result in unbounded estimates if two nodes always share the same state in the data set.
`maxit, trace, hessian, ...`	Arguments passed to `optim`
`parlist`	user-specified starting values (optional).

a list with the following elements:

`alpha`	A vector of estimated univariate potentials ("intercepts"). Larger values support higher probabilities for the corresponding node.
`beta`	A symmetric matrix of bivariate potentials ("interaction strengths"). Larger values support higher probabilities for the corresponding pair of nodes.
`prior`	The prior object that was used during model fitting.
`opt`	The list returned by `optim`.

# Simulate a random binary matrix with 1000 observations of five binary variables
m = matrix(rbinom(5000, size = 1, prob = .5), nrow = 1000, ncol = 5)

fit = rosalia(m)

fit$alpha
fit$beta