April 2, 2014
Charlotte's RStudio preferences tips:
If you have large objects in memory you don't want to be saving unneccessary copies and loading them each time you start R. It also forces you to write reproducible code.
This intro is no substitute for reading the introduction vignette. Read it!
We'll probably also assign the databases and memory vignettes in future, but there's no harm in taking a look at them early.
dplyr is a package for data manipulation and exploration
It's designed to be fast and avoid unneccessary memory consumption.
It's a useful mental model, which means it reduces cognitive effort: you expend some energy learning dplyr now, in exchange for less brain power required in future for routine data analysis tasks.
Master dplyr on data.frames and move seamlessly to databases.
dplyr is not plyr!
To run code later:
library(dplyr) library(hflights) hflights_df <- tbl_df(hflights)
An action on a data.frame, results in a data.frame
First argument is always a data.frame, remaining arguments specify the action, (no need for $):
verb(hflights_df, ...)
| Verb | Action | Example |
|---|---|---|
| filter | subset rows | filter(hflights_df, Dest == "PDX") |
| select | subset columns | select(hflights_df, ArrDelay, UniqueCarrier) |
| arrange | reorder rows | arrange(hflights_df, desc(ArrDelay)) |
| mutate | add new columns | mutate(hflights_df, more15 = ArrDelay > 15) |
| summarise | reduce to a single row | summarise(hflights_df, avg_delay = mean(ArrDelay, na.rm = TRUE)) |
Apply to a data.frame to define a "grouping" of rows based on the levels of one or more columns. The verbs know about groups:
group_by first:
carriers <- group_by(hflights_df, UniqueCarrier)
then use a verb:
summarise(carriers, median_delay = median(ArrDelay, na.rm = TRUE))
Just a matter of learning to translate questions into a sequence verbs and grouping operations. Then writing the code is easy.
Which day in 2011 had the most delays?
do, do some function within each group. For example, let's you do things like fit a regression model to each group and keep the results in a list.n(), n_distinct(), first(), last(), nth(),?ranking, lag(), lead()