@markfairbanks, @mgirlich, and @eutwt are now dtplyr authors in recognition of their significant and sustained contributions. Along with @eutwt, they supplied the bulk of the improvements in this release!
dtplyr gains translations for many more tidyr verbs:
drop_na() (@markfairbanks, #194)complete() (@markfairbanks, #225)expand() (@markfairbanks, #225)fill() (@markfairbanks, #197)pivot_longer() (@markfairbanks, #204)replace_na() (@markfairbanks, #202)nest() (@mgirlich, #251)separate() (@markfairbanks, #269)tally() gains a translation (@mgirlich, #201).
ifelse() is mapped to fifelse() (@markfairbanks, #220).
slice() helpers (slice_head(), slice_tail(), slice_min(), slice_max() and slice_sample()) now accept negative values for n and prop.
across() defaults to everything() when .cols isn’t provided (@markfairbanks, #231), and handles named selections (@eutwt #293). It ˜ow handles .fns arguments in more forms (@eutwt #288):
function(x) x + 1~ 1arrange(dt, desc(col)) is translated to dt[order(-col)] in order to take advantage of data.table’s fast order (@markfairbanks, #227).
count() applied to data.tables no longer breaks when dtplyr is loaded (@mgirlich, #201).
case_when() supports use of T to specify the default (#272).
filter() errors for named input, e.g. filter(dt, x = 1) (@mgirlich, #267) and works for negated logical columns (@mgirlich, @211).
group_by() ungroups when no grouping variables are specified (@mgirlich, #248), and supports inline mutation like group_by(dt, y = x) (@mgirlich, #246).
if_else() named arguments are translated to the correct arguments in data.table::fifelse() (@markfairbanks, #234). if_else() supports .data and .env pronouns (@markfairbanks, #220).
if_any() and if_all() default to everything() when .cols isn’t provided (@eutwt, #294).
intersect()/union()/union_all()/setdiff() convert data.table inputs to lazy_dt() (#278).
lag()/lead() are translated to shift().
lazy_dt() keeps groups (@mgirlich, #206).
left_join() produces the same column order as dplyr (@markfairbanks, #139).
left_join(), right_join(), full_join(), and inner_join() perform a cross join for by = character() (@mgirlich, #242).
left_join(), right_join(), and inner_join() are always translated to the [.data.table equivalent. For simple merges the translation gets a bit longer but thanks to the simpler code base it helps to better handle names in by and duplicated variables names produced in the data.table join (@mgirlich, #222).
mutate() and transmute() work when called without variables (@mgirlich, #248).
mutate() gains new experimental arguments .before and .after that allow you to control where the new columns are placed (to match dplyr 1.0.0) (@eutwt #291).
mutate() can modify grouping columns (instead of creating another column with the same name) (@mgirlich, #246).
n_distinct() is translated to uniqueN().
tally() and count() follow the dplyr convention of creating a unique name if the default output name (n) already exists (@eutwt, #295).
pivot_wider() names the columns correctly when names_from is a numeric column (@mgirlich, #214).
pull() supports the name argument (@mgirlich, #263).
slice() no longer returns excess rows (#10).
slice_*() functions after group_by() are faster (@mgirlich, #216).
slice_max() works when ordering by a character column (@mgirlich, #218).
summarise() supports the .groups argument (@mgirlich, #245).
summarise(), tally(), and count() can change the value of a grouping variables (@eutwt, #295).
transmute() doesn’t produce duplicate columns when assigning to the same variable (@mgirlich, #249). It correctly flags grouping variables so they selected (@mgirlich, #246).
ungroup() removes variables in ... from grouping (@mgirlich, #253).
All verbs now have (very basic) documentation pointing back to the dplyr generic, and providing a (very rough) description of the translation accompanied with a few examples.
Passing a data.table to a dplyr generic now converts it to a lazy_dt(), making it a little easier to move between data.table and dplyr syntax.
dtplyr has been bought up to compatibility with dplyr 1.0.0. This includes new translations for:
across(), if_any(), if_all() (#154).
count() (#159).
relocate() (@smingerson, #162).
rename_with() (#160)
slice_min(), slice_max(), slice_head(), slice_tail(), and slice_sample() (#174).
And rename() and select() now support dplyr 1.0.0 tidyselect syntax (apart from predicate functions which can’t easily work on lazily evaluated data tables).
We have begun the process of adding translations for tidyr verbs beginning with pivot_wider() (@markfairbanks, #189).
compute() now creates an intermediate assignment within the translation. This will generally have little impact on performance but it allows you to use intermediate variables to simplify complex translations.
case_when() is now translated to fcase() (#190).
cur_data() (.SD), cur_group() (.BY), cur_group_id() (.GRP), and cur_group_rows() (.I`) are now tranlsated to their data.table equivalents (#166).
filter() on grouped data nows use a much faster translation using on .I rather than .SD (and requiring an intermediate assignment) (#176). Thanks to suggestion from @myoung3 and @ColeMiller1.
Translation of individual expressions:
x[[1]] is now translated correctly.
Anonymous functions are now preserved (@smingerson, #155)
Environment variables used in the i argument of [.data.table are now correctly inlined when not in the global environment (#164).
T and F are correctly translated to TRUE and FALSE (#140).
Grouped filter, mutate, and slice no longer affect ordering of output (#178).
as_tibble() gains a .name_repair argument (@markfairbanks).
as.data.table() always calls [] so that the result will print (#146).
print.lazy_dt() shows total rows, and grouping, if present.
group_map() and group_walk() are now translated (#108).
Better handling for .data and .env pronouns (#138).
dplyr verbs now work with NULL inputs (#129).
joins do better job at determining output variables in the presence of duplicated outputs (#128). When joining based on different variables in x and y, joins consistently preserve column from x, not y (#137).
lazy_dt() objects now have a useful glimpse() method (#132).
group_by() now has an arrange parameter which, if set to FALSE, sets the data.table translation to use by rather than keyby (#85).
rename() now works without data.table attached, as intended (@michaelchirico, #123).
dtplyr has been re-licensed as MIT (#165).
Converted from eager approach to lazy approach. You now must use lazy_dt() to begin a translation pipeline, and must use collect(), as.data.table(), as.data.frame(), or as_tibble() to finish the translation and actually perform the computation (#38).
This represents a complete overhaul of the package replacing the eager evaluation used in the previous releases. This unfortunately breaks all existing code that used dtplyr, but frankly the previous version was extremely inefficient so offered little of data.table’s impressive speed, and was used by very few people.
dtplyr provides methods for data.tables that warning you that they use the data frame implementation and you should use lazy_dt() (#77)
Joins now pass ... on to data.table’s merge method (#41).
ungroup() now copies its input (@christophsax, #54).
mutate() preserves grouping (@christophsax, #17).
if_else() and coalesce() are mapped to data.table’s fifelse() and fcoalesce() respectively (@michaelchirico, #112).
Maintenance release for CRAN checks.
inner_join(), left_join(), right_join(), and full_join(): new suffix argument which allows you to control what suffix duplicated variable names receive, as introduced in dplyr 0.5 (#40, @christophsax).
Joins use extended merge.data.table() and the on argument, introduced in data.table 1.9.6. Avoids copy and allows joins by different keys (#20, #21, @christophsax).
distinct() gains .keep_all argument (#30, #31).
Slightly improve test coverage (#6).
Install devtools from GitHub on Travis (#32).
Joins return data.table. Right and full join are now implemented (#16, #19).
Remove warnings from tests (#4).
Extracted from dplyr at revision e5f2952923028803.