condmixt.init {condmixt}R Documentation

Conditional mixture parameter initial values

Description

Neural network weights are randomly initialized uniformly over the range [-0.9/sqrt(k),0.9/sqrt(k)] where k is the number of inputs to the neuron. This ensures that the hidden units will not be saturated and that training should proceed properly. In addition, if the dependent data Y is provided, the biases will be initialized according to the initial parameters of an unconditional mixture computed on the dependent data.

Usage

condhparetomixt.init(d, h, m, y = NULL)
condhparetomixt.dirac.init(d, h, m, y = NULL)
condgaussmixt.init(d,h,m,y=NULL)
condgaussmixt.dirac.init(d,h,m,y=NULL)
condlognormixt.init(d,h,m,y=NULL)
condlognormixt.dirac.init(d,h,m,y=NULL)
condbergamixt.init(d,h,y=NULL)

Arguments

d

dimension of input x to neural network

h

number of hidden unit

m

number of components

y

optional, dependent one-dimensional data

Details

If the argument y is provided, an unconditional mixture with the same type of components will be initialized on y. These initial unconditional parameters are then used to give more appropriate initial values to the biases of the neural network.

Value

A vector of neural network parameters for the given number of hidden units and number of components and specific conditional mixture formulation : hybrid Pareto components (condhparetomixt.init), hybrid Pareto components + discrete dirac component at zero (condhparetomixt.dirac.init), Gaussian components (condgaussmixt.init), Gaussian components + discrete dirac component at zero (condgaussmixt.dirac.init), Log-Normal components (condlognormixt.init), Log-Normal components + discrete dirac component at zero (condlognormixt.dirac.init) and the Bernoulli-Gamma two-component mixture (condbergamixt.init).

Author(s)

Julie Carreau

References

Carreau, J. and Bengio, Y. (2009), A Hybrid Pareto Mixture for Conditional Asymmetric Fat-Tailed Distributions, 20, IEEE Transactions on Neural Networks

Nabney, I. (2002) NetLab : Algorithms for Pattern Recognition, Springer

Williams, M.P. (1998) Modelling Seasonality and Trends in Daily Rainfall Data, 10, Advances in Neural Information and Processing Systems

See Also

hparetomixt.init, gaussmixt.init

Examples

n <- 200
x <- runif(n) # x is a random uniform variate
# y depends on x through the parameters of the Frechet distribution
y <- rfrechet(n,loc = 3*x+1,scale = 0.5*x+0.001,shape=x+1)

plot(x,y,pch=22)
# 0.99 quantile of the generative distribution
qgen <- qfrechet(0.99,loc = 3*x+1,scale = 0.5*x+0.001,shape=x+1)
points(x,qgen,pch="*",col="orange")

h <- 2 # number of hidden units
m <- 4 # number of components

# initialize a conditional mixture with hybrid Pareto components
thetainit <- condhparetomixt.init(1,h,m,y)

[Package condmixt version 1.1 Index]