lossRandom: Loss function

In heike/dbData: Database Access for Sufficient Statistics of Data Graphics

Description

Based on a given data set, loss is computed from comparing the original data to the data binned at the specified level. Loss is computed as sum of squared differences between all values at the original data and the binned data. Data that are "in beween" two bins are randomly assigned to one bin or the other by a bernoulli trial with success probability equal to the distance between the bin's center and the point divided by the binning level. Percent loss compares loss to the total sum of squares in the original data.

Usage

  lossRandom(data, binning, newData = FALSE)

Arguments

d1	data frame
binning	vector of binwidths
newData	binned data available?

Examples

connect <- dbConnect(dbDriver("MySQL"), user="2009Expo",
password="R R0cks", port=3306, dbname="baseball",
host="headnode.stat.iastate.edu")
pitch <- new("dataDB", co=connect, table="Pitching")
d1 <- dbData(pitch, vars=c( "G", "SO"))
qplot(G,SO, fill=log10(Freq), data=d1, geom="tile")+scale_fill_gradient2()
lossRandom(d1, binning=c(2,5))
lossRandom(d1, binning=c(1,5))
d2 <- dbData(pitch, vars=c( "G", "SO"), binwidth=c(1,5))
qplot(G,SO, fill=log10(Freq), data=d2, geom="tile")+scale_fill_gradient2()
lossRandom(d1, binning=c(1,10))
lossRandom(d2, binning=c(1,2))
d2 <- dbData(pitch, vars=c( "G", "SO"), binwidth=c(1,10))
qplot(G,SO, fill=log10(Freq), data=d2, geom="tile")+scale_fill_gradient2()
## some more exploration of loss
bins <- expand.grid(x=c(1:10), y=c(1:10))
losses <- mdply(bins, function(x,y) lossRandom(data=d1, binning=c(x, y)))
qplot(x, percent.loss, group=y, data=losses, geom="line")
qplot(y, percent.loss, group=x, data=losses, geom="line")
d2 <- dbData(pitch, vars=c( "G", "SO"), binwidth=c(10,10))

heike/dbData documentation built on May 17, 2019, 3:23 p.m.