get_log_perms {perms}

get_log_perms

Description

Computes log permanents associated with simulated latent variables. Each row of the S x n matrix X contains a random sample of size n from the data model. If there is only a single covariate, then the observed data are represented as (t,y), where t is the observed values of the covariate and y is the vector of indicator variables. If there are more covariates or the problem is phrased as binary classification (see Section 5 in [1]), then t is an S x n matrix since the threshold values change in each iteration. The function returns a vector of log permanents corresponding to each sample in X.

Usage

get_log_perms(X, tt, y, debug = FALSE, parallel = TRUE, num_cores = NULL)

Arguments

X

A matrix of dimension S x n, in which each row contains a sample from the data model.

tt

Either: A vector of length n containing the observed values of the covariate, Or: A matrix of dimension S x n (if there are several covariates).

y

A vector of length n indicating whether x_i<=t_i for each i in the observed data.

debug

If TRUE, debug information is printed.

parallel

If TRUE, computation is run on several cores

num_cores

(Optional) Specifies the number of cores to use if parallel = TRUE

Value

Numpy array of log permanents,each element associated to the corresponding row in X. A zero valued permanent is indicated by a NA value.

References

[1] Christensen, D (2023). Inference for Bayesian nonparametric models with binary response data via permutation counting. Bayesian Analysis, Advance online publication, DOI: 10.1214/22-BA1353.

Examples

library(perms)
set.seed(1996)
n = 100
t = seq(0, 1, length.out=n)
y = c(rep(0, n/2), rep(1, n/2))
S = 200
X = matrix(runif(n*S),nrow = S, ncol = n)

log_perms = get_log_perms(X, t, y, debug = FALSE, parallel = FALSE, num_cores = NULL)

num_nonzero_perms = sum(!is.na(log_perms))
num_nonzero_perms

log_ML = get_log_ML(log_perms, n, FALSE)
log_ML