remove_small_sample_states: Reassign low sample states to close states

In LICORS: Light Cone Reconstruction of States - Predictive State Estimation From Spatio-Temporal Data

Description

This function removes small sample states by reassigning points in those state to nearby states.

This can become necessary when in an iterative algorithm (like mixed_LICORS) the weights start moving away from e.g. state j. At some point the effective sample size of state j (sum of column \mathbf{W}_j) is so small that state-conditional estimates (mean, variance, kernel density estimate, etc.) can not be obtained accurately anymore. Then it is good to remove state j and reassign its samples to other (close) states.

Usage

remove_small_sample_states(weight.matrix, min)

Arguments

weight.matrix	N \times K weight matrix
min	minimum effective sample size to stay in the weight matrix

Examples

set.seed(10)
WW <- matrix(c(rexp(1000, 1/10), runif(1000)), ncol = 5, byrow = FALSE)
WW <- normalize(WW)
colSums(WW)
remove_small_sample_states(WW, 20)

LICORS documentation built on May 1, 2019, 10:13 p.m.