split

A common usage of `split()` is to split a data frame by a factor. ```{r} data(PlantGrowth) split(PlantGrowth, PlantGrowth$group) ``` You can also split vectors. ```{r} with(PlantGrowth, split(weight, group)) ``` To split by numeric ranges, combine `split()` with `cut()`. ```{r} split(PlantGrowth, cut(PlantGrowth$weight, 3:7)) ``` `drop = TRUE` drops unused elements of length zero in the resulting list. ```{r} (f <- factor( sample(c("a", "e", "i", "o", "u"), 20, replace = TRUE), levels = letters )) split(1:20, f) split(1:20, f, drop = TRUE) ``` You can also split by passing a numeric index. ```{r} split(1:10, 1:2) ``` A more useful example is to split a data frame into a list of one-row data frames suitable for use with [`lapply()`](https://www.rdocumentation.org/packages/base/topics/lapply). ```{r} head(plant_list <- split(PlantGrowth, seq_len(nrow(PlantGrowth)))) sapply(plant_list, function(x) x$weight ^ 2) ``` To split by multiple factors, pass them in a list. ```{r} with(CO2, split(CO2, list(Plant, Type, Treatment), drop = TRUE)) ``` A common data analysis pattern is to split a dataset into groups, apply an action to each group, then combine those results together. See [The Split-Apply-Combine Strategy for Data Analysis](https://www.jstatsoft.org/article/view/v040i01) by Hadley Wickham (2011). ```{r} # split plant_weights_by_group <- with(PlantGrowth, split(weight, group)) # apply mean_plant_weights_by_group <- lapply(plant_weights_by_group, mean) # combine unlist(mean_plant_weights_by_group) ``` We can speed this up: [`sapply()`](https://www.rdocumentation.org/packages/base/topics/lapply) does the last two steps together. Even better, [`tapply()`](https://www.rdocumentation.org/packages/base/topics/tapply) does all three. ```{r} with(PlantGrowth, tapply(weight, group, mean)) ``` Another common split-apply-combine problem would be calculate 'z-scores' by group (scaled to mean zero, variance one). This time, the answer is the same length as the input. ```{r} plant_weights_by_group <- with(PlantGrowth, split(weight, group)) scaled_plant_weights_by_group <- lapply(plant_weights_by_group, scale) unlist(scaled_plant_weights_by_group) ``` The quick version uses [`ave()`](https://www.rdocumentation.org/packages/stats/topics/ave) instead of [`tapply()`](https://www.rdocumentation.org/packages/base/topics/tapply) ```{r} with(PlantGrowth, ave(weight, group, FUN = scale)) ``` Occasionally, you may want to reverse to splitting process. ```{r} plant_weights_by_group <- with(PlantGrowth, split(weight, group)) unsplit(plant_weights_by_group, PlantGrowth$group) ``` For vectors it is usually easier to use [`unlist()`](https://www.rdocumentation.org/packages/base/topics/unlist), since you don't need to remember what the splitter was. ```{r} plant_weights_by_group <- with(PlantGrowth, split(weight, group)) unlist(plant_weights_by_group) ``` Similarly, for data frames, [`do.call()`](https://www.rdocumentation.org/packages/base/topics/do.call) + [`rbind()`](https://www.rdocumentation.org/packages/base/topics/rbind) (or [`dplyr::bind_all()`](https://www.rdocumentation.org/packages/dplyr/topics/bind_all)) is usually easier. ```{r} plants_by_group <- split(PlantGrowth, PlantGrowth$group) do.call(rbind, plants_by_group) ``` There are many alternatives for solving split-apply-combine problems. `dplyr`'s [`group_by_()`](https://www.rdocumentation.org/packages/dplyr/topics/group_by) is its equivalent of `split()`. ```{r} library(dplyr) # tapply() equivalent PlantGrowth %>% group_by_(~ group) %>% summarize_(mean_weight = ~ mean(weight)) # ave() equivalent PlantGrowth %>% group_by_(~ group) %>% mutate_(scaled_weight = ~ scale(weight)) ``` Similarly, `data.table` has a `by` argument. ```{r} library(data.table) # tapply() equivalent as.data.table(PlantGrowth)[, .(mean_weight = mean(weight)), by = group] # ave() equivalent pg <- as.data.table(PlantGrowth)[, scaled_weight := scale(weight), by = group] print(pg) ```