Dplyr on Data.Table, am I Really Using Data.Table

dplyr on data.table, am I really using data.table?

There is no straightforward/simple answer because the philosophies of both these packages differ in certain aspects. So some compromises are unavoidable. Here are some of the concerns you may need to address/consider.

Operations involving `i` (== `filter()` and `slice()` in dplyr)

Assume DT with say 10 columns. Consider these data.table expressions:

DT[a > 1, .N]                    ## --- (1)
DT[a > 1, mean(b), by=.(c, d)]   ## --- (2)

(1) gives the number of rows in DT where column a > 1. (2) returns mean(b) grouped by c,d for the same expression in i as (1).

Commonly used dplyr expressions would be:

DT %>% filter(a > 1) %>% summarise(n())                        ## --- (3) 
DT %>% filter(a > 1) %>% group_by(c, d) %>% summarise(mean(b)) ## --- (4)

Clearly, data.table codes are shorter. In addition they are also more memory efficient¹. Why? Because in both (3) and (4), filter() returns rows for all 10 columns first, when in (3) we just need the number of rows, and in (4) we just need columns b, c, d for the successive operations. To overcome this, we have to select() columns apriori:

DT %>% select(a) %>% filter(a > 1) %>% summarise(n()) ## --- (5)
DT %>% select(a,b,c,d) %>% filter(a > 1) %>% group_by(c,d) %>% summarise(mean(b)) ## --- (6)

It is essential to highlight a major philosophical difference between the two packages:

In data.table, we like to keep these related operations together, and that allows to look at the j-expression (from the same function call) and realise there's no need for any columns in (1). The expression in i gets computed, and .N is just sum of that logical vector which gives the number of rows; the entire subset is never realised. In (2), just column b,c,d are materialised in the subset, other columns are ignored.

But in dplyr, the philosophy is to have a function do precisely one thing well. There is (at least currently) no way to tell if the operation after filter() needs all those columns we filtered. You'll need to think ahead if you want to perform such tasks efficiently. I personally find it counter-intutitive in this case.

Note that in (5) and (6), we still subset column a which we don't require. But I'm not sure how to avoid that. If filter() function had an argument to select the columns to return, we could avoid this issue, but then the function will not do just one task (which is also a dplyr design choice).

Sub-assign by reference

dplyr will never update by reference. This is another huge (philosophical) difference between the two packages.

For example, in data.table you can do:

DT[a %in% some_vals, a := NA]

which updates column a by reference on just those rows that satisfy the condition. At the moment dplyr deep copies the entire data.table internally to add a new column. @BrodieG already mentioned this in his answer.

But the deep copy can be replaced by a shallow copy when FR #617 is implemented. Also relevant: dplyr: FR#614. Note that still, the column you modify will always be copied (therefore tad slower / less memory efficient). There will be no way to update columns by reference.

Other functionalities

In data.table, you can aggregate while joining, and this is more straightfoward to understand and is memory efficient since the intermediate join result is never materialised. Check this post for an example. You can't (at the moment?) do that using dplyr's data.table/data.frame syntax.
data.table's rolling joins feature is not supported in dplyr's syntax as well.
We recently implemented overlap joins in data.table to join over interval ranges (here's an example), which is a separate function foverlaps() at the moment, and therefore could be used with the pipe operators (magrittr / pipeR? - never tried it myself).
But ultimately, our goal is to integrate it into [.data.table so that we can harvest the other features like grouping, aggregating while joining etc.. which will have the same limitations outlined above.
Since 1.9.4, data.table implements automatic indexing using secondary keys for fast binary search based subsets on regular R syntax. Ex: DT[x == 1] and DT[x %in% some_vals] will automatically create an index on the first run, which will then be used on successive subsets from the same column to fast subset using binary search. This feature will continue to evolve. Check this gist for a short overview of this feature.
From the way filter() is implemented for data.tables, it doesn't take advantage of this feature.
A dplyr feature is that it also provides interface to databases using the same syntax, which data.table doesn't at the moment.

So, you will have to weigh in these (and probably other points) and decide based on whether these trade-offs are acceptable to you.

HTH

(1) Note that being memory efficient directly impacts speed (especially as data gets larger), as the bottleneck in most cases is moving the data from main memory onto cache (and making use of data in cache as much as possible - reduce cache misses - so as to reduce accessing main memory). Not going into details here.

data.table vs dplyr memory use revisited

UPDATE: Following @jangorecki's suggestion, I redid the analysis using the cgmemtime shell utility. The numbers are far closer — even with multithreading enabled — and data.table now edges out dplyr w.r.t to .high-water RSS+CACHE memory usage.

dplyr

$ ./cgmemtime Rscript ~/mem-comp-dplyr.R
Child user:    0.526 s
Child sys :    0.033 s
Child wall:    0.455 s
Child high-water RSS                    :     128952 KiB
Recursive and acc. high-water RSS+CACHE :     118516 KiB

data.table

$ ./cgmemtime Rscript ~/mem-comp-dt.R
Child user:    0.510 s
Child sys :    0.056 s
Child wall:    0.464 s
Child high-water RSS                    :     129032 KiB
Recursive and acc. high-water RSS+CACHE :     118320 KiB

Bottom line: Accurately measuring memory usage from within R is complicated.

I'll leave my original answer below because I think it still has value.

ORIGINAL ANSWER:

Okay, so in the process of writing this out I realised that data.table's default multi-threading behaviour appears to be the major culprit. If I re-run the latter chunk, but this time turn of multi-threading, the two results are much more comparable:

library(bench)
library(dplyr, warn.conflicts = FALSE)
library(data.table, warn.conflicts = FALSE)
set.seed(123)
setDTthreads(1) ## TURN OFF MULTITHREADING

DT = data.table(x = rep(1:10, times = 1e5),
                y = sample(LETTERS[1:10], 10e5, replace = TRUE),
                z = rnorm(1e6))

DT[x > 7, mean(z), by = y]
#>     y            V1
#>  1: F -0.0056834238
#>  2: I -0.0016755202
#>  3: J  0.0066061660
#>  4: G -0.0034436348
#>  5: B -0.0070242788
#>  6: E -0.0070462070
#>  7: H  0.0005525803
#>  8: D -0.0043024627
#>  9: A -0.0033609302
#> 10: C  0.0029146372

bench::bench_process_memory()
#> current     max 
#>   589MB   612MB

^{Created on 2020-04-22 by the reprex package (v0.3.0)}

Still, I'm surprised that they're this close. The data.table memory performance actually gets comparably worse if I try with a larger data set — despite using a single thread — which makes me suspicious that I'm still not measuring memory usage correctly...

data.table vs dplyr: can one do something well the other can't or does poorly?

We need to cover at least these aspects to provide a comprehensive answer/comparison (in no particular order of importance): Speed, Memory usage, Syntax and Features.

My intent is to cover each one of these as clearly as possible from data.table perspective.

Note: unless explicitly mentioned otherwise, by referring to dplyr, we refer to dplyr's data.frame interface whose internals are in C++ using Rcpp.

The data.table syntax is consistent in its form - DT[i, j, by]. To keep i, j and by together is by design. By keeping related operations together, it allows to easily optimise operations for speed and more importantly memory usage, and also provide some powerful features, all while maintaining the consistency in syntax.

1. Speed

Quite a few benchmarks (though mostly on grouping operations) have been added to the question already showing data.table gets faster than dplyr as the number of groups and/or rows to group by increase, including benchmarks by Matt on grouping from 10 million to 2 billion rows (100GB in RAM) on 100 - 10 million groups and varying grouping columns, which also compares pandas. See also updated benchmarks, which include Spark and pydatatable as well.

On benchmarks, it would be great to cover these remaining aspects as well:

Grouping operations involving a subset of rows - i.e., DT[x > val, sum(y), by = z] type operations.
Benchmark other operations such as update and joins.
Also benchmark memory footprint for each operation in addition to runtime.

2. Memory usage

Operations involving filter() or slice() in dplyr can be memory inefficient (on both data.frames and data.tables). See this post.
Note that Hadley's comment talks about speed (that dplyr is plentiful fast for him), whereas the major concern here is memory.
data.table interface at the moment allows one to modify/update columns by reference (note that we don't need to re-assign the result back to a variable).
```
 # sub-assign by reference, updates 'y' in-place
 DT[x >= 1L, y := NA]
```
But dplyr will never update by reference. The dplyr equivalent would be (note that the result needs to be re-assigned):
```
 # copies the entire 'y' column
 ans <- DF %>% mutate(y = replace(y, which(x >= 1L), NA))
```
A concern for this is referential transparency. Updating a data.table object by reference, especially within a function may not be always desirable. But this is an incredibly useful feature: see this and this posts for interesting cases. And we want to keep it.
Therefore we are working towards exporting shallow() function in data.table that will provide the user with both possibilities. For example, if it is desirable to not modify the input data.table within a function, one can then do:
```
 foo <- function(DT) {
     DT = shallow(DT)          ## shallow copy DT
     DT[, newcol := 1L]        ## does not affect the original DT 
     DT[x > 2L, newcol := 2L]  ## no need to copy (internally), as this column exists only in shallow copied DT
     DT[x > 2L, x := 3L]       ## have to copy (like base R / dplyr does always); otherwise original DT will 
                               ## also get modified.
 }
```
By not using shallow(), the old functionality is retained:
```
 bar <- function(DT) {
     DT[, newcol := 1L]        ## old behaviour, original DT gets updated by reference
     DT[x > 2L, x := 3L]       ## old behaviour, update column x in original DT.
 }
```
By creating a shallow copy using shallow(), we understand that you don't want to modify the original object. We take care of everything internally to ensure that while also ensuring to copy columns you modify only when it is absolutely necessary. When implemented, this should settle the referential transparency issue altogether while providing the user with both possibilties.
Also, once shallow() is exported dplyr's data.table interface should avoid almost all copies. So those who prefer dplyr's syntax can use it with data.tables.
But it will still lack many features that data.table provides, including (sub)-assignment by reference.
Aggregate while joining:
Suppose you have two data.tables as follows:
```
 DT1 = data.table(x=c(1,1,1,1,2,2,2,2), y=c("a", "a", "b", "b"), z=1:8, key=c("x", "y"))
 #    x y z
 # 1: 1 a 1
 # 2: 1 a 2
 # 3: 1 b 3
 # 4: 1 b 4
 # 5: 2 a 5
 # 6: 2 a 6
 # 7: 2 b 7
 # 8: 2 b 8
 DT2 = data.table(x=1:2, y=c("a", "b"), mul=4:3, key=c("x", "y"))
 #    x y mul
 # 1: 1 a   4
 # 2: 2 b   3
```
And you would like to get sum(z) * mul for each row in DT2 while joining by columns x,y. We can either:
- 1. aggregate DT1 to get sum(z), 2) perform a join and 3) multiply (or)
    data.table way
    DT1[, .(z = sum(z)), keyby = .(x,y)][DT2][, z := z*mul][]
    dplyr equivalent
    DF1 %>% group_by(x, y) %>% summarise(z = sum(z)) %>%
    right_join(DF2) %>% mutate(z = z * mul)
- 1. do it all in one go (using by = .EACHI feature):
    DT1[DT2, list(z=sum(z) * mul), by = .EACHI]
What is the advantage?
- We don't have to allocate memory for the intermediate result.
- We don't have to group/hash twice (one for aggregation and other for joining).
- And more importantly, the operation what we wanted to perform is clear by looking at j in (2).
Check this post for a detailed explanation of by = .EACHI. No intermediate results are materialised, and the join+aggregate is performed all in one go.
Have a look at this, this and this posts for real usage scenarios.
In dplyr you would have to join and aggregate or aggregate first and then join, neither of which are as efficient, in terms of memory (which in turn translates to speed).
Update and joins:
Consider the data.table code shown below:
```
 DT1[DT2, col := i.mul]
```
adds/updates DT1's column col with mul from DT2 on those rows where DT2's key column matches DT1. I don't think there is an exact equivalent of this operation in dplyr, i.e., without avoiding a *_join operation, which would have to copy the entire DT1 just to add a new column to it, which is unnecessary.
Check this post for a real usage scenario.

To summarise, it is important to realise that every bit of optimisation matters. As Grace Hopper would say, Mind your nanoseconds!

3. Syntax

Let's now look at syntax. Hadley commented here:

Data tables are extremely fast but I think their concision makes it harder to learn and code that uses it is harder to read after you have written it ...

I find this remark pointless because it is very subjective. What we can perhaps try is to contrast consistency in syntax. We will compare data.table and dplyr syntax side-by-side.

We will work with the dummy data shown below:

DT = data.table(x=1:10, y=11:20, z=rep(1:2, each=5))
DF = as.data.frame(DT)

Basic aggregation/update operations.
```
 # case (a)
 DT[, sum(y), by = z]                       ## data.table syntax
 DF %>% group_by(z) %>% summarise(sum(y)) ## dplyr syntax
 DT[, y := cumsum(y), by = z]
 ans <- DF %>% group_by(z) %>% mutate(y = cumsum(y))

 # case (b)
 DT[x > 2, sum(y), by = z]
 DF %>% filter(x>2) %>% group_by(z) %>% summarise(sum(y))
 DT[x > 2, y := cumsum(y), by = z]
 ans <- DF %>% group_by(z) %>% mutate(y = replace(y, which(x > 2), cumsum(y)))

 # case (c)
 DT[, if(any(x > 5L)) y[1L]-y[2L] else y[2L], by = z]
 DF %>% group_by(z) %>% summarise(if (any(x > 5L)) y[1L] - y[2L] else y[2L])
 DT[, if(any(x > 5L)) y[1L] - y[2L], by = z]
 DF %>% group_by(z) %>% filter(any(x > 5L)) %>% summarise(y[1L] - y[2L])
```
- data.table syntax is compact and dplyr's quite verbose. Things are more or less equivalent in case (a).
- In case (b), we had to use filter() in dplyr while summarising. But while updating, we had to move the logic inside mutate(). In data.table however, we express both operations with the same logic - operate on rows where x > 2, but in first case, get sum(y), whereas in the second case update those rows for y with its cumulative sum.
  This is what we mean when we say the DT[i, j, by] form is consistent.
- Similarly in case (c), when we have if-else condition, we are able to express the logic "as-is" in both data.table and dplyr. However, if we would like to return just those rows where the if condition satisfies and skip otherwise, we cannot use summarise() directly (AFAICT). We have to filter() first and then summarise because summarise() always expects a single value.
  While it returns the same result, using filter() here makes the actual operation less obvious.
  It might very well be possible to use filter() in the first case as well (does not seem obvious to me), but my point is that we should not have to.
Aggregation / update on multiple columns
```
 # case (a)
 DT[, lapply(.SD, sum), by = z]                     ## data.table syntax
 DF %>% group_by(z) %>% summarise_each(funs(sum)) ## dplyr syntax
 DT[, (cols) := lapply(.SD, sum), by = z]
 ans <- DF %>% group_by(z) %>% mutate_each(funs(sum))

 # case (b)
 DT[, c(lapply(.SD, sum), lapply(.SD, mean)), by = z]
 DF %>% group_by(z) %>% summarise_each(funs(sum, mean))

 # case (c)
 DT[, c(.N, lapply(.SD, sum)), by = z]     
 DF %>% group_by(z) %>% summarise_each(funs(n(), mean))
```
- In case (a), the codes are more or less equivalent. data.table uses familiar base function lapply(), whereas dplyr introduces *_each() along with a bunch of functions to funs().
- data.table's := requires column names to be provided, whereas dplyr generates it automatically.
- In case (b), dplyr's syntax is relatively straightforward. Improving aggregations/updates on multiple functions is on data.table's list.
- In case (c) though, dplyr would return n() as many times as many columns, instead of just once. In data.table, all we need to do is to return a list in j. Each element of the list will become a column in the result. So, we can use, once again, the familiar base function c() to concatenate .N to a list which returns a list.
Note: Once again, in data.table, all we need to do is return a list in j. Each element of the list will become a column in result. You can use c(), as.list(), lapply(), list() etc... base functions to accomplish this, without having to learn any new functions.
You will need to learn just the special variables - .N and .SD at least. The equivalent in dplyr are n() and .

Joins

dplyr provides separate functions for each type of join where as data.table allows joins using the same syntax DT[i, j, by] (and with reason). It also provides an equivalent merge.data.table() function as an alternative.

 setkey(DT1, x, y)

 # 1. normal join
 DT1[DT2]            ## data.table syntax
 left_join(DT2, DT1) ## dplyr syntax

 # 2. select columns while join    
 DT1[DT2, .(z, i.mul)]
 left_join(select(DT2, x, y, mul), select(DT1, x, y, z))

 # 3. aggregate while join
 DT1[DT2, .(sum(z) * i.mul), by = .EACHI]
 DF1 %>% group_by(x, y) %>% summarise(z = sum(z)) %>% 
     inner_join(DF2) %>% mutate(z = z*mul) %>% select(-mul)

 # 4. update while join
 DT1[DT2, z := cumsum(z) * i.mul, by = .EACHI]
 ??

 # 5. rolling join
 DT1[DT2, roll = -Inf]
 ??

 # 6. other arguments to control output
 DT1[DT2, mult = "first"]
 ??

Some might find a separate function for each joins much nicer (left, right, inner, anti, semi etc), whereas as others might like data.table's DT[i, j, by], or merge() which is similar to base R.
However dplyr joins do just that. Nothing more. Nothing less.
data.tables can select columns while joining (2), and in dplyr you will need to select() first on both data.frames before to join as shown above. Otherwise you would materialiase the join with unnecessary columns only to remove them later and that is inefficient.
data.tables can aggregate while joining (3) and also update while joining (4), using by = .EACHI feature. Why materialse the entire join result to add/update just a few columns?
data.table is capable of rolling joins (5) - roll forward, LOCF, roll backward, NOCB, nearest.
data.table also has mult = argument which selects first, last or all matches (6).
data.table has allow.cartesian = TRUE argument to protect from accidental invalid joins.

Once again, the syntax is consistent with DT[i, j, by] with additional arguments allowing for controlling the output further.

do()...

dplyr's summarise is specially designed for functions that return a single value. If your function returns multiple/unequal values, you will have to resort to do(). You have to know beforehand about all your functions return value.

 DT[, list(x[1], y[1]), by = z]                 ## data.table syntax
 DF %>% group_by(z) %>% summarise(x[1], y[1]) ## dplyr syntax
 DT[, list(x[1:2], y[1]), by = z]
 DF %>% group_by(z) %>% do(data.frame(.$x[1:2], .$y[1]))

 DT[, quantile(x, 0.25), by = z]
 DF %>% group_by(z) %>% summarise(quantile(x, 0.25))
 DT[, quantile(x, c(0.25, 0.75)), by = z]
 DF %>% group_by(z) %>% do(data.frame(quantile(.$x, c(0.25, 0.75))))

 DT[, as.list(summary(x)), by = z]
 DF %>% group_by(z) %>% do(data.frame(as.list(summary(.$x))))

.SD's equivalent is .
In data.table, you can throw pretty much anything in j - the only thing to remember is for it to return a list so that each element of the list gets converted to a column.
In dplyr, cannot do that. Have to resort to do() depending on how sure you are as to whether your function would always return a single value. And it is quite slow.

Once again, data.table's syntax is consistent with DT[i, j, by]. We can just keep throwing expressions in j without having to worry about these things.

Have a look at this SO question and this one. I wonder if it would be possible to express the answer as straightforward using dplyr's syntax...

To summarise, I have particularly highlighted several instances where dplyr's syntax is either inefficient, limited or fails to make operations straightforward. This is particularly because data.table gets quite a bit of backlash about "harder to read/learn" syntax (like the one pasted/linked above). Most posts that cover dplyr talk about most straightforward operations. And that is great. But it is important to realise its syntax and feature limitations as well, and I am yet to see a post on it.

data.table has its quirks as well (some of which I have pointed out that we are attempting to fix). We are also attempting to improve data.table's joins as I have highlighted here.

But one should also consider the number of features that dplyr lacks in comparison to data.table.

4. Features

I have pointed out most of the features here and also in this post. In addition:

fread - fast file reader has been available for a long time now.
fwrite - a parallelised fast file writer is now available. See this post for a detailed explanation on the implementation and #1664 for keeping track of further developments.
Automatic indexing - another handy feature to optimise base R syntax as is, internally.
Ad-hoc grouping: dplyr automatically sorts the results by grouping variables during summarise(), which may not be always desirable.
Numerous advantages in data.table joins (for speed / memory efficiency and syntax) mentioned above.
Non-equi joins: Allows joins using other operators <=, <, >, >= along with all other advantages of data.table joins.
Overlapping range joins was implemented in data.table recently. Check this post for an overview with benchmarks.
setorder() function in data.table that allows really fast reordering of data.tables by reference.
dplyr provides interface to databases using the same syntax, which data.table does not at the moment.
data.table provides faster equivalents of set operations (written by Jan Gorecki) - fsetdiff, fintersect, funion and fsetequal with additional all argument (as in SQL).
data.table loads cleanly with no masking warnings and has a mechanism described here for [.data.frame compatibility when passed to any R package. dplyr changes base functions filter, lag and [ which can cause problems; e.g. here and here.

Finally:

On databases - there is no reason why data.table cannot provide similar interface, but this is not a priority now. It might get bumped up if users would very much like that feature.. not sure.
On parallelism - Everything is difficult, until someone goes ahead and does it. Of course it will take effort (being thread safe).
- Progress is being made currently (in v1.9.7 devel) towards parallelising known time consuming parts for incremental performance gains using OpenMP.

Dangers of mixing [tidyverse] and [data.table] syntax in R?

I came across the same problem on a few occasions, which led me to avoid mixing dplyr with data.table syntax, as I didn't take the time to find out the reason. So thanks for providing a MRE.

Looks like dplyr::arrange is interfering with data.table auto-indexing :

index will be used when subsetting dataset with ==
or %in%
on a single variable
by default if index for a variable is not present on filtering, it is automatically created and used
indexes are lost if you change the order of data
you can check if you are using index with options(datatable.verbose=TRUE)

If we explicitely set auto-indexing :

library(dplyr); 
library(data.table)

DT <- fread(
"iso3c  country income
MOZ Mozambique  LIC
ZMB Zambia  LMIC
ALB Albania UMIC
MOZ Mozambique  LIC
ZMB Zambia  LMIC
ALB Albania UMIC")
codes <- c("ALB", "ZMB")

options(datatable.auto.index = TRUE)

DT <- distinct(DT) %>%   as.data.table()

# Index creation because %in% is used for the first time
DT[iso3c %in% codes,verbose=T]
#> Creating new index 'iso3c'
#> Creating index iso3c done in ... forder.c received 3 rows and 3 columns
#> forder took 0 sec
#> 0.060s elapsed (0.060s cpu) 
#> Optimized subsetting with index 'iso3c'
#> forder.c received 2 rows and 1 columns
#> forder took 0 sec
#> x is already ordered by these columns, no need to call reorder
#> i.iso3c has same type (character) as x.iso3c. No coercion needed.
#> on= matches existing index, using index
#> Starting bmerge ...
#> bmerge done in 0.000s elapsed (0.000s cpu) 
#> Constructing irows for '!byjoin || nqbyjoin' ... 0.000s elapsed (0.000s cpu) 
#> Reordering 2 rows after bmerge done in ... forder.c received a vector type 'integer' length 2
#> 0 secs
#>    iso3c country income
#> 1:   ZMB  Zambia   LMIC
#> 2:   ALB Albania   UMIC

# Index mixed up by arrange
DT <- DT %>% arrange(iso3c) %>% as.data.table()

# this is wack because data.table possibly still uses the old index whereas row/references were rearranged:
DT[iso3c %in% codes,verbose=T]
#> Optimized subsetting with index 'iso3c'
#> forder.c received 2 rows and 1 columns
#> forder took 0 sec
#> x is already ordered by these columns, no need to call reorder
#> i.iso3c has same type (character) as x.iso3c. No coercion needed.
#> on= matches existing index, using index
#> Starting bmerge ...
#> bmerge done in 0.000s elapsed (0.000s cpu) 
#> Constructing irows for '!byjoin || nqbyjoin' ... 0.000s elapsed (0.000s cpu)
#>    iso3c country income
#> 1:   ALB Albania   UMIC

# this works because (...) prevents the parser to use auto-index
DT[(iso3c %in% codes)]
#>    iso3c country income
#> 1:   ALB Albania   UMIC
#> 2:   ZMB  Zambia   LMIC

Dplyr on Data.Table, am I Really Using Data.Table