A Pipe version with recommended practice

README.md

A Pipe version of Zev Ross.

An alternative version of timelyportfolio's and smbache's.

If you are not familiar with pipeR, you can read the still-in-progress pipeR tutorial. The best practices of pipeR's Pipe():

• Keep side effects as small as possible, use it only when necessary.
• Separate Pipes into concentrated pieces.
• Don't avoid necessary intermediate variables.
• The purpose is always readability, not throwing everything into a big pipe.

After all, using pipeline does not automatically make your code better!.

code.R

library(pipeR)
library(dplyr)
library(ggplot2)
library(mgcv)

# get the data ready
pdata <- Pipe("http://zevross.com/blog/wp-content/uploads/2014/08/chicago-nmmaps.csv")$
  read.csv(as.is = TRUE)$
  mutate(date = as.Date(date))$
  filter(date > as.Date("1996-12-31"))$
  mutate(year = substring(date,1,4))

# get a gam in Pipe
pgam <- pdata$
  gam(formula = o3~s(temp), family = gaussian())

# make a plot
pgam$plot(residuals = TRUE, main="Yuck, not a nice plot")

# get the new data in Pipe
pnewdata <- pdata$
  with(data.frame(temp = seq(min(temp),max(temp),length=300)))

# make some predictions with new data
pred <- pgam$
  predict(type = "terms", newdata = pnewdata$value, se.fit=TRUE)$
  value

# draw a blank board
pnewdata$.(temp)$
  plot(pred$fit, type="n", lwd=3, xlim=c(-3,90), ylim=c(-20,30),
    main="Ahhh, definitely better",
    ylab=sprintf("s(temp, %f)",pgam$with(edf[-1])$sum()$round(2)$value))

# draw the confidence grey polygon
pnewdata$with(c(temp, rev(temp)))$
  polygon(with(pred, c(fit+1.96*se.fit, rev(fit-1.96*se.fit))),
    col="grey", border=NA)

# draw fit lines
pnewdata$.(temp)$lines(pred$fit, lwd=2)

# draw points
pdata$.(temp)$
  points(pgam$predict(type="terms")$value + pgam$residuals()$value,
    pch=16, col=rgb(0, 0, 1, 0.25))

# draw rug
pdata$.(temp)$rug()

more-compact.R

library(pipeR)
library(dplyr)
library(ggplot2)
library(mgcv)

# get the data ready
pdata <- Pipe("http://zevross.com/blog/wp-content/uploads/2014/08/chicago-nmmaps.csv")$
  read.csv(as.is = TRUE)$
  mutate(date = as.Date(date))$
  filter(date > as.Date("1996-12-31"))$
  mutate(year = substring(date,1,4))

# get a gam in Pipe
pgam <- pdata$gam(formula = o3~s(temp), family = gaussian())
pgam$plot(residuals = TRUE, main="Yuck, not a nice plot")

# get the new data in Pipe
pnewdata <- pdata$with(data.frame(temp = seq(min(temp),max(temp),length=300)))

# make some predictions with new data
pred <- pgam$
  predict(type = "terms", newdata = pnewdata$value, se.fit=TRUE)$
  value

pnewdata$.(temp)$.({
  plot(., pred$fit, type="n", lwd=3, xlim=c(-3,90), ylim=c(-20,30),
    main="Ahhh, definitely better",
    ylab=sprintf("s(temp, %f)",pgam$with(edf[-1])$sum()$round(2)$value))
  polygon(c(.,rev(.)),
    with(pred, fit+1.96*se.fit, rev(fit-1.96*se.fit)),
    col="grey", border=NA)
  lines(., pred$fit, lwd=2)
})

pdata$.(temp)$.({
  points(.,
    pgam$predict(type="terms")$value + pgam$residuals()$value,
    pch=16, col=rgb(0, 0, 1, 0.25))
  rug(.)
})