README.md

In alexgenin/rollply: Moving-Window Add-on for 'plyr'

rollply

Rollply is an R package built on top on plyr's **ply functions to make moving-window based computations easier. Just provide a function that works on a data.frame, rollply will take care of building the appropriate subsets and apply that function to each of them. It will combine the results into a data.frame at the end of the computation.

Some packages out there contain functions that compute moving-window statistics (e.g. TTR::SMA), but were not flexible enough for my use and required specific time-series classes. Rollply works on generic data.frames and builds upon plyr's advantages (parallelism, progress report, etc.), but also on its weaknesses (it's slow).

In short, it allows writing something like this:

library(ggplot2)
library(plyr)
library(rollply)
# Generate data
dat <- data.frame(time=seq.int(1000),
                  position=cumsum(rnorm(1000,0,10)))

rollav <- rollply(dat, ~ time, wdw.size=10,
                  summarise, position=mean(position))

ggplot(NULL,aes(time,position)) +
  geom_point(data=dat) +
  geom_line(color='red', data=rollav)

Rollply supports groups and 2D moving windows too:

# Generate three 2D random walks
dat <- ddply(data.frame(person=c('françois','nicolas','jacques')), ~ person,
             summarise,
              time=seq.int(1000),
              x=cumsum(rnorm(1000,0,1)),
              y=cumsum(rnorm(1000,0,1)))

# Smoothed trajectory over ten time-steps
rollav <- rollply(dat, ~ time | person, wdw.size=10, grid_npts=1000,
                  summarise, x=mean(x), y=mean(y))

ggplot(dat,aes(x,y,color=person)) +
  geom_point(alpha=.5, shape='+') +
  geom_path(data=rollav)

# Where did people spend their time ?
# we fix the grid across groups
fixed_grid <- build_grid_squaretile(dat[ ,c('x','y')], 5000)
rollav <- rollply(dat, ~ x + y | person, wdw.size=2, grid=fixed_grid,
                  summarise, time.spent=length(time))

ggplot(subset(rollav, time.spent>0)) +
  geom_point(aes(x,y, color=person, size=time.spent)) +
  facet_grid(~person)

Installation

The best way to install rollply is from github:

r if ( !require(devtools) ) { install.packages('devtools') library(devtools) } install_github('alexgenin/rollply') library(rollply)

The CRAN version has been deprecated as I do not have currently the time to fix the CRAN-related issues. It may come back at a later stage.

alexgenin/rollply documentation built on May 10, 2019, 8:51 a.m.