Portability  nonportable (requires STM) 

Stability  experimental 
Maintainer  gnn.github@gmail.com 
The GSOM Algorithm can be split up in multiple sequentially executed
s. Each of these phases makes a certain number of passes over
the inputs. While doing so each Phase
modifies a given phase
according to a certain set of specified parameters.
This module contains the definition of the Lattice
type, a few default
instances and the functions needed to run a single Phase
or to
phase
a sequence of run
s.
Phase
 data Phase = Phase {
 passes :: Int
 neighbourhoodSize :: Int
 learningRate :: LearningRate
 kernel :: Kernel
 grow :: Bool
 spreadFactor :: Double
 type Phases = [Phase]
 defaultFirst :: Phase
 defaultSecond :: Phase
 defaultThird :: Phase
 defaults :: Phases
 phase :: Phase > Lattice > Inputs > IO Lattice
 run :: Phases > Lattice > Inputs > IO Lattice
 data Kernel
 data LearningRate
 = Linear Double
  InverseAge Double
Documentation
This datatype encapsulates all the parameters needed to be known to run one phase of the GSOM algorithm.
Phase  

The three default phases of the GSOM algorithm. They all use the bubble kernel and a linear learning rate decrease.
The default first phase is the only growing phase. It makes 5
passes over the input, uses an initial learning rate of 0.1 and
a starting neighbourhood size of 3. The
is set
to 0.1.
spreadFactor
The default for the second phase is a smoothing phase making 50
passes over the input vectors with a learning rate of 0.05 and an
initial neighbourhood size of 2. Since there is no node growth the
is ignored and thus set to 0.
spreadFactor
The default for the third and last phase is a smoothing phase making
50 passes over the input vectors with a learning rate of 0.01 and
an initial neighbourhood size of 1. Since there is no node growth the
is ignored and thus set to 0.
spreadFactor
phase :: Phase > Lattice > Inputs > IO LatticeSource
will update the given phase
parameters inputslattice
by
executing one phase of the GSOM algorithm with the given inputs
and parameters
.
run :: Phases > Lattice > Inputs > IO LatticeSource
Since a complete run of the GSOM algorithm means running a number of
this is usually the main function used.
Phases
run phases lattice inputs
runs the GSOM algorithm by running the
phases
in the order specified, each time making passes over inputs
and using the produced
to as an argument to the next phase.
The initial Lattice
, Lattice
lattice
may be constructed with the
and the newRandom
functions.
newCentered
Bubble  The bubble kernel is essentially the identity, i.e. it has no effect. 
Gaussian  Let

data LearningRate Source
Linear Double  The linear learning rate reduction function. If you supply it with
the initial learning rate

InverseAge Double  The inverse time learning rate reduction function. Given an initial
learning rate of
