devianceCategorical: Function to calculate deviance for model predictions assuming...

In EdwinGraham/EdNet: Train deep neural networks for classification and regression

Description

This function calculates a deviance measure for model predictions assuming a Categorical distribution.

Usage

devianceCategorical(Y, Y_hat)

Arguments

Y	a numeric matrix of observations (0s and 1s) where columns indicate categories and rows indicate observations. Should be exactly one 1 per row.
Y_hat	a numeric matrix of predictions (between 0 and 1) for Y (must have same dimensions as Y.) Row sums should all equal 1.

Value

a numeric vector.

Author(s)

Edwin Graham <[e-mail]>

Examples

# n <- 1000
# nCat <- 10
# 
# # Random probabilities normalised by row
# true_logistic <- matrix(rnorm(n*nCat), ncol = nCat)
# true_probabilities <- exp(true_logistic)/rowSums(exp(true_logistic))
# 
# # Generate observations
# observed <- t(apply(true_probabilities, 1, cumsum))
# observed <- 1*(observed > runif(n))
# observed <- 1*t(apply(observed, 1, cumsum)==1)
# 
# # Generate predictions
# predicted <- true_logistic + matrix(rnorm(n*nCat, sd=0.1), ncol=nCat)
# predicted <- exp(predicted)/rowSums(exp(predicted))
# 
# plot(observed, predicted)
# 
# devs <- devianceCategorical(observed, predicted)
# sum(devs)

EdwinGraham/EdNet documentation built on May 6, 2019, 12:22 p.m.