Description Usage Arguments Value Note See Also Examples
sTrainSeq
is supposed to perform sequential training algorithm.
It requires three inputs: a "sMap" or "sInit" object, input data, and a
"sTrain" object specifying training environment. The training is
implemented iteratively, each training cycle consisting of: i) randomly
choose one input vector; ii) determine the winner hexagon/rectangle
(BMH) according to minimum distance of codebook matrix to the input
vector; ii) update the codebook matrix of the BMH and its neighbors via
updating formula (see "Note" below for details). It also returns an
object of class "sMap".
1 |
sMap |
an object of class "sMap" or "sInit" |
data |
a data frame or matrix of input data |
sTrain |
an object of class "sTrain" |
seed |
an integer specifying the seed |
verbose |
logical to indicate whether the messages will be displayed in the screen. By default, it sets to TRUE for display |
an object of class "sMap", a list with following components:
nHex
: the total number of hexagons/rectanges in the grid
xdim
: x-dimension of the grid
ydim
: y-dimension of the grid
r
: the hypothetical radius of the grid
lattice
: the grid lattice
shape
: the grid shape
coord
: a matrix of nHex x 2, with each row corresponding
to the coordinates of a hexagon/rectangle in the 2D map grid
ig
: the igraph object
init
: an initialisation method
neighKernel
: the training neighborhood kernel
codebook
: a codebook matrix of nHex x ncol(data), with
each row corresponding to a prototype vector in input high-dimensional
space
call
: the call that produced this result
Updating formula is: m_i(t+1) = m_i(t) + α(t)*h_{wi}(t)*[x(t)-m_i(t)], where
t denotes the training time/step
i and w stand for the hexagon/rectangle i and the winner BMH w, respectively
x(t) is an input vector randomly choosen (from the input data) at time t
m_i(t) and m_i(t+1) are respectively the prototype vectors of the hexagon i at time t and t+1
α(t) is the learning rate at time t. There are three types of learning rate functions:
For "linear" function, α(t)=α_0*(1-t/T)
For "power" function, α(t)=α_0*(0.005/α_0)^{t/T}
For "invert" function, α(t)=α_0/(1+100*t/T)
Where α_0 is the initial learing rate (typically, α_0=0.5 at "rough" stage, α_0=0.05 at "finetune" stage), T is the length of training time/step (often being set to input data length, i.e., the total number of rows)
h_{wi}(t) is the neighborhood kernel, a non-increasing function of i) the distance d_{wi} between the hexagon/rectangle i and the winner BMH w, and ii) the radius δ_t at time t. There are five kernels available:
For "gaussian" kernel, h_{wi}(t)=e^{-d_{wi}^2/(2*δ_t^2)}
For "cutguassian" kernel, h_{wi}(t)=e^{-d_{wi}^2/(2*δ_t^2)}*(d_{wi} ≤ δ_t)
For "bubble" kernel, h_{wi}(t)=(d_{wi} ≤ δ_t)
For "ep" kernel, h_{wi}(t)=(1-d_{wi}^2/δ_t^2)*(d_{wi} ≤ δ_t)
For "gamma" kernel, h_{wi}(t)=1/Γ(d_{wi}^2/(4*δ_t^2)+2)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 | # 1) generate an iid normal random matrix of 100x10
data <- matrix( rnorm(100*10,mean=0,sd=1), nrow=100, ncol=10)
# 2) from this input matrix, determine nHex=5*sqrt(nrow(data))=50,
# but it returns nHex=61, via "sHexGrid(nHex=50)", to make sure a supra-hexagonal grid
sTopol <- sTopology(data=data, lattice="hexa", shape="suprahex")
# 3) initialise the codebook matrix using "uniform" method
sI <- sInitial(data=data, sTopol=sTopol, init="uniform")
# 4) define trainology at "rough" stage
sT_rough <- sTrainology(sMap=sI, data=data, algorithm="sequential",
stage="rough")
# 5) training at "rough" stage
sM_rough <- sTrainSeq(sMap=sI, data=data, sTrain=sT_rough)
# 6) define trainology at "finetune" stage
sT_finetune <- sTrainology(sMap=sI, data=data, algorithm="sequential",
stage="finetune")
# 7) training at "finetune" stage
sM_finetune <- sTrainSeq(sMap=sM_rough, data=data, sTrain=sT_rough)
|
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.