stochastic_search: Stochastic Search

Description Usage Arguments Details

View source: R/stochastic_search.R

Description

An stochasic algorithm for optimisation

Usage

1
2
3
stochastic_search(dim_param, perf_fun, loss_fun = ls_loss, target_perf,
  max_iter = 100, tol = 0, curiosity = 1, block_num, block_size,
  lambda = 0, reg_fun = zero_reg, param)

Arguments

dim_param

integer; the dimension of the parameters.

perf_fun

function; the performance function mapping parameters to an outcome, e.g. a curve.

loss_fun

function; the loss function of two outcomes.

target_perf

function; the performance to be matched, e.g. a target data curve.

max_iter

integer; maximum number of iterations.

tol

numeric; tolerence for the loss functions.

curiosity

numeric; search parameter.

block_num

integer.

block_size

integer.

lambda

numeric; regularisation parameter.

reg_fun

function; regularisation funciton, see 'zero_reg', 'L1_reg', 'L2_reg'.

param

numeric vector; parameter from previous run of the function.

Details

Denote the dimension of the parameters by d. If the stochastic search didn't improve in 5*d block iterations, then it will increase the curiosity by 10 folds. On the other hand, all improvements are associated with a 0.001 decrease curiosity. This is to create oscillatary behaviour of the curiosity parameter.


kcf-jackson/wandeRer documentation built on May 20, 2019, 8:15 a.m.