gmum.r (version 0.2.1)

OptimizedGNG: Constructor of Optimized GrowingNeuralGas object.

Description

Construct simplified and optimized GNG object. Can be used to train offline, or online. Data dimensionality shouldn't be too big, if it is consider using dimensionality reduction techniques.

Usage

OptimizedGNG(x = NULL, labels = c(), beta = 0.99, alpha = 0.5, max.nodes = 1000, eps.n = 6e-04, eps.w = 0.05, max.edge.age = 200, train.online = FALSE, max.iter = 200, dim = 0, min.improvement = 0.001, lambda = 200, verbosity = 0, seed = -1, value.range = c(0, 1))

Arguments

x

Passed data (matrix of data.frame) for offline training

labels

Every example can be associated with labels that are added to nodes later. By default empty

beta

Decrease the error variables of all node nodes by this fraction (forgetting rate). Default 0.99

alpha

Decrease the error variables of the nodes neighboring to the newly inserted node by this fraction. Default 0.5

max.nodes

Maximum number of nodes (after reaching this size it will continue running, but new noes won't be added)

eps.n

Strength of adaptation of neighbour node. Default 0.0006

eps.w

Strength of adaptation of winning node. Default 0.05

max.edge.age

Maximum edge age. Decrease to increase speed of change of graph topology. Default 200

train.online

If used will run in online fashion. Default FALSE

max.iter

If training offline will stop if exceedes max.iter iterations. Default 200

dim

Used for training online, specifies dataset example dimensionality

min.improvement

Used for offline (default) training. Controls stopping criterion, decrease if training stops too early. Default 1e-3

lambda

New vertex is added every lambda iterations. Default 200

verbosity

How verbose should the process be, as integer from $[0,6]$, default: 0

seed

Seed for internal randomization

value.range

All example features should be in this range, required for optimized version of the algorithm. Default (0,1)

Examples

## Not run: 
# # Train online optimizedGNG. All values in this dataset are in the range (-4.3, 4.3)
# X <- gng.preset.sphere(100)
# gng <- OptimizedGNG(train.online = TRUE, value.range=c(min(X), max(X)), dim=3, max.nodes=20)
# insertExamples(gng, X)
# run(gng)
# Sys.sleep(10)
# pause(gng)
# ## End(Not run)