问题
I am trying to add a time-varying predictor to a long-form dataframe using reshape2::melt but I was wondering if there was a faster way to do it.
Here is the toy data in wide form. There are three measures of an outcome variable (session1, session2, and session3) taken at different visits/time points. The duration between these three visits is different for each participant and ultimately I would like to factor these differences into a model.
id <- 1:10
group <- rep(c("A", "B"), times = 5)
session1 <- rnorm(10, 5, 1)
session2 <- rnorm(10, 3, 1)
session3 <- rnorm(10, 7, 2)
time1 <- rep(0, 10)
time2 <- rnorm(10, 24, 0.5)
time3 <- rnorm(10, 48, 0.5)
df <- data.frame(id, group, session1, session2, session3, time1, time2, time3)
Now I want to convert into a long-form dataframe. I use reshape2::melt. I can create this either around the scores like so:
library(reshape2)
dfLong <- melt(df, measure.vars = c("session1", "session2", "session3"), var = "session", value.name = "score")
Or I can create it around the time values.
dfLong2 <- melt(df, measure.vars = c("time1", "time2", "time3"), var = "time", value.name = "timeOut")
But I can't do both without doing the melt twice and performing some sort of opertation like this
dfLong$timeOut <- dfLong2$timeOut
Ultimately I would like the dataframe to look something like this
dfLong$time <- rep(c("time1", "time2", "time3"), each = 10)
dfLong <- dfLong[,which(names(dfLong) %in% c("id", "group", "time", "session", "score", "timeOut"))]
dfLong
Is there any way to melt two sets of columns at once?
回答1:
We can use data.table
library(data.table)
res = melt(setDT(df), measure = patterns("^session", "^time"),
value.name = c("session", "time"))
You can setDF(res) to revert to a data.frame if you don't want to learn how to work with data.tables right now.
来源：https://stackoverflow.com/questions/37884946/melting-two-sets-of-columns-from-wide-to-long-form-using-reshape2