lossMean: 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 mean of the data in the new bin. Percent loss compares loss to the total sum of squares in the original data.

Usage

  lossMean(data, binning, newData = FALSE)

Arguments

d1	data frame
binning	vector of binwidths

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()
lossMean(d1, binning=c(2,5))
lossMean(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()
lossMean(d1, binning=c(1,10))
lossMean(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) lossMean(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.