How to use data.table within functions and loops?
While assessing the utility of data.table (vs. dplyr), a critical factor is the ability to use it within functions and loops.
For this, I've modified the code snippet used in this post: data.table vs dplyr: can one do something well the other can't or does poorly? so that, instead of hard-coded dataset variables names ("cut" and "price" variables of "diamonds" dataset), it becomes dataset-agnostic - cut-n-paste ready for the use inside any function or a loop (when we don't know column names in advance).
This is the original code:
library(data.table)
dt <- data.table(ggplot2::diamonds)
dt[cut != "Fair", .(mean(price),.N), by = cut]
This is its dataset-agnostic equivalent:
dt <- data.table(diamonds)
nVarGroup <- 2 #"cut"
nVarMeans <- 7 #"price"
strGroupConditions <- levels(dt[[nVarGroup]])[-1] # "Good" "Very Good" "Premium" "Ideal"
strVarGroup <- names(dt)[nVarGroup]
strVarMeans <- names(dt)[nVarMeans]
qAction <- quote(mean(get(strVarMeans))) #! w/o get() it does not work!
qGroup <- quote(get(strVarGroup) %in% strGroupConditions) #! w/o get() it does not work!
dt[eval(qGroup), .(eval(qAction), .N), by = strVarGroup]
Note (Thanks to reply below): if you need to change variable value by reference, you need to use (), not get(), as shown below:
strVarToBeReplaced <- names(dt)[1]
dt[eval(qGroup), (strVarToBeReplaced) := eval(qAction), by = strGroup][]
Now: you can cut-n-paste the following code for all your looping needs:
for(nVarGroup in 2:4) # Grouped by several categorical values...
for(nVarMeans in 5:10) { # ... get means of all numerical parameters
strGroupConditions <- levels(dt[[nVarGroup]])[-1]
strVarGroup <- names(dt)[nVarGroup]
strVarMeans <- names(dt)[nVarMeans]
qAction <- quote(mean(get(strVarMeans)))
qGroup <- quote(get(strVarGroup) %in% strGroupConditions)
p <- dt[eval(qGroup), .(AVE=eval(qAction), COUNT=.N), by = strVarGroup]
print(sprintf("nVaGroup=%s, nVarMeans=%s: ", strVarGroup, strVarMeans))
print(p)
}
My first question:
The code above, while enabling the required functional/looping needs, appears quite convoluted. - It uses different multiple (possibly non-consistent) non-intuitive tricks such combination of (), get(), quote()/eval(), [[]]). Seems too many for a such straightforward need...
Is there another better way of accessing and modifying data.tables values in loops? Perhaps with on=, lapply/.SD/.SDcols?
Please share your ideas below. This discussion aims to supplement and consolidate related bits from other posts (such as listed here: How can one work fully generically in data.table in R with column names in variables). Eventually, it would be great to create a dedicated vignette for using data.table within functions and loops.
The second question:
Is dplyr easier for this purpose? - For this question however, I've set a separate post: Is dplyr easier than data.table to be used within functions and loops?.
The OP has asked for a dataset-agnostic equivalent for grouping and aggregating.
With development version 1.10.5, data.table has gained new Grouping Sets functions: rollup(), cube(), and groupingsets() which allow aggregation on various grouping levels at once producing sub-totals and grand total.
The added level of abstraction can be used for a dataset-agnostic approach. The subtotals which are computed using a double nested for loop in OP's example can be produced as well by
library(data.table) # version 1.10.5 required
dt = data.table(ggplot2::diamonds)
groupingsets(dt, c(lapply(.SD, mean), list(COUNT = .N)),
by = names(dt)[2:4], .SDcols = 5:10, id = FALSE,
sets = as.list(names(dt)[2:4]))
cut color clarity depth table price x y z COUNT 1: Ideal NA NA 61.70940 55.95167 3457.542 5.507451 5.520080 3.401448 21551 2: Premium NA NA 61.26467 58.74610 4584.258 5.973887 5.944879 3.647124 13791 3: Good NA NA 62.36588 58.69464 3928.864 5.838785 5.850744 3.639507 4906 4: Very Good NA NA 61.81828 57.95615 3981.760 5.740696 5.770026 3.559801 12082 5: Fair NA NA 64.04168 59.05379 4358.758 6.246894 6.182652 3.982770 1610 6: NA E NA 61.66209 57.49120 3076.752 5.411580 5.419029 3.340689 9797 7: NA I NA 61.84639 57.57728 5091.875 6.222826 6.222730 3.845411 5422 8: NA J NA 61.88722 57.81239 5323.818 6.519338 6.518105 4.033251 2808 9: NA H NA 61.83685 57.51781 4486.669 5.983335 5.984815 3.695965 8304 10: NA F NA 61.69458 57.43354 3724.886 5.614961 5.619456 3.464446 9542 11: NA G NA 61.75711 57.28863 3999.136 5.677543 5.680192 3.505021 11292 12: NA D NA 61.69813 57.40459 3169.954 5.417051 5.421128 3.342827 6775 13: NA NA SI2 61.77217 57.92718 5063.029 6.401370 6.397826 3.948478 9194 14: NA NA SI1 61.85304 57.66254 3996.001 5.888383 5.888256 3.639845 13065 15: NA NA VS1 61.66746 57.31515 3839.455 5.572178 5.581828 3.441007 8171 16: NA NA VS2 61.72442 57.41740 3924.989 5.657709 5.658859 3.491478 12258 17: NA NA VVS2 61.66378 57.02499 3283.737 5.218454 5.232118 3.221465 5066 18: NA NA VVS1 61.62465 56.88446 2523.115 4.960364 4.975075 3.061294 3655 19: NA NA I1 62.73428 58.30378 3924.169 6.761093 6.709379 4.207908 741 20: NA NA IF 61.51061 56.50721 2864.839 4.968402 4.989827 3.061659 1790
So, we do not have to know the names of the columns. However, we have to specify which columns to group by and which columns to aggregate.
This might not be the most data.table-like or the fastest solution but I would streamline the code in this particular loop as follows:
for(nVarGroup in 2:4) { # Grouped by several categorical values...
for(nVarMeans in 5:10) { # ... get means of all numerical parameters
strGroupConditions <- levels(dt[[nVarGroup]])[-1]
strVarGroup <- names(dt)[nVarGroup]
strVarMeans <- names(dt)[nVarMeans]
# qAction <- quote(mean(get(strVarMeans)))
# qGroup <- quote(get(strVarGroup) %in% strGroupConditions)
# p <- dt[eval(qGroup), .(AVE = eval(qAction), COUNT = .N), by = strVarGroup]
setkeyv(dt, strVarGroup)
p <- dt[strGroupConditions, .(AVE = lapply(.SD, mean), COUNT = .N), by = strVarGroup,
.SDcols = strVarMeans]
print(sprintf("nVaGroup = %s, nVarMeans = %s", strVarGroup, strVarMeans))
print(p)
}
}
I've left the old code as comment for reference.
qAction is replaced by using lapply(.SD, mean) together with the .SDcols parameter.
qGroup for subsetting rows is replaced by the combination of setting a key and providing the vector of desired values as i parameter.
In case of a more complex subsetting expression I would try use non-equi (or conditional) joins using the on= syntax.
Or, follow Matt Dowle's advice to create one expression to be evaluated, "similar to constructing a dynamic SQL statement to send to a server".
Matt suggested to create a helper function
EVAL <- function(...) eval(parse(text = paste0(...)), envir = parent.frame(2))
which can be combined with the "quasi-perl type string interpolation of fn$ from the gsubfn package to improve the readability of the EVAL solution" as suggested by G. Grothendieck.
With this, the code for the loop becomes eventually:
EVAL <- function(...) eval(parse(text = paste0(...)), envir = parent.frame(2))
library(gsubfn)
for(nVarGroup in 2:4) { # Grouped by several categorical values...
for(nVarMeans in 5:10) { # ... get means of all numerical parameters
strGroupConditions = levels(dt[[nVarGroup]])[-1]
strVarGroup = names(dt)[nVarGroup]
strVarMeans = names(dt)[nVarMeans]
p <- fn$EVAL("dt[$strVarGroup %in% strGroupConditions, .(AVE=mean($strVarMeans), COUNT=.N), by = strVarGroup]" )
print(sprintf("nVaGroup = %s, nVarMeans = %s", strVarGroup, strVarMeans))
print(p)
}
}
Now, the data.table statement looks pretty much like a "native" statement except that $strVarGroup and $strVarMeans is used where the contents of variables is referenced.
With version 1.1.0 (CRAN release on 2016-08-19), the stringr package has gained a string interpolation function str_interp() which is an alternative to the gsubfn package here.
With str_interp(), the central statement in the for loop would become
p <- EVAL(stringr::str_interp(
"dt[${strVarGroup} %in% strGroupConditions, .(AVE=mean(${strVarMeans}), COUNT=.N), by = strVarGroup]"
))
and the call to library(gsubfn) could be removed.