Log-likelihood of LICORS model

Description

Computes the average log-likelihood \frac{1}{N} \ell(\mathbf{W}; \mathcal{D}) as a function of the weight matrix \mathbf{W} and the predictive state distributions P(X = x \mid S = s_j) \approx P(X = x \mid \mathbf{W}_j) for all j = 1, …, K. See References.

Usage

1
2
compute_LICORS_loglik(weight.matrix, pdfs.FLC, lambda = 0, penalty = "entropy", q = 2, 
    base = exp(1))

Arguments

weight.matrix

N \times K weight matrix

pdfs.FLC

an N \times K matrix containing the estimates of all K FLC densities evaluated at all N sample FLCs.

lambda

regularization parameter. Default: lambda=0 (penalty and q will be ignored in this case).

penalty

type of penalty: c("entropy", "1-Lq", "lognorm"). Default: "entropy"

base

logarithm base for the "entropy" penalty. Default: base = 2. Any other real number is allowed; if base = "num.states" then it will internally assign it base = ncol(weight.matrix).

q

exponent for L_q norm.

Want to suggest features or report bugs for rdrr.io? Use the GitHub issue tracker.