randnet (version 0.7)

LSM.PGD: estimates inner product latent space model by projected gradient descent

Description

estimates inner product latent space model by projected gradient descent from the paper of Ma et al. (2020).

Usage

LSM.PGD(A, k,step.size=0.3,niter=500,trace=0)

Value

a list of

Z

latent positions

alpha

individual parameter alpha as in the paper

Phat

esitmated probability matrix

obj

the objective of the gradient method

Arguments

A

adjacency matrix

k

the dimension of the latent position

step.size

step size of gradient descent

niter

maximum number of iterations

trace

if trace > 0, the objective will be printed out after each iteration

Author

Tianxi Li and Can M. Le
Maintainer: Tianxi Li

Details

The method is based on the gradient descent of Ma et al (2020), with initialization of the universal singular value thresholding as discussed there. The parameter identifiability constraint is the same as in the paper.

References

Z. Ma, Z. Ma, and H. Yuan. Universal latent space model fitting for large networks with edge covariates. Journal of Machine Learning Research, 21(4):1-67, 2020.

See Also

DCSBM.estimate

Examples

dt <- RDPG.Gen(n=600,K=2,directed=TRUE)


A <- dt$A


fit <- LSM.PGD(A,2,niter=50)