fmatch: Fast match() replacement

View source: R/fastmatch.R

fmatchR Documentation

Fast match() replacement


fmatch is a faster version of the built-in match() function. It is slightly faster than the built-in version because it uses more specialized code, but in addition it retains the hash table within the table object such that it can be re-used, dramatically reducing the look-up time especially for large tables.

Although fmatch can be used separately, in general it is also safe to use: match <- fmatch since it is a drop-in replacement. Any cases not directly handled by fmatch are passed to match with a warning.

fmatch.hash is identical to fmatch but it returns the table object with the hash table attached instead of the result, so it can be used to create a table object in cases where direct modification is not possible.

%fin% is a version of the built-in %in% function that uses fmatch instead of match().


fmatch(x, table, nomatch = NA_integer_, incomparables = NULL)
fmatch.hash(x, table, nomatch = NA_integer_, incomparables = NULL)
x %fin% table



values to be matched


values to be matched against


the value to be returned in the case when no match is found. It is coerced to integer.


a vector of values that cannot be matched. Any value other than NULL will result in a fall-back to match without any speed gains.


See match for the purpose and details of the match function. fmatch is a drop-in replacement for the match function with the focus on performance. incomparables are not supported by fmatch and will be passed down to match.

The first match against a table results in a hash table to be computed from the table. This table is then attached as the ".match.hash" attribute of the table so that it can be re-used on subsequent calls to fmatch with the same table.

The hashing algorithm used is the same as the match function in R, but it is re-implemented in a slightly different way to improve its performance at the cost of supporting only a subset of types (integer, real and character). For any other types fmatch falls back to match (with a warning).


fmatch: A vector of the same length as x - see match for details.

fmatch.hash: table, possibly coerced to match the type of x, with the hash table attached.

%fin%: A logical vector the same length as x - see %in% for details.


fmatch modifies the table by attaching an attribute to it. It is expected that the values will not change unless that attribute is dropped. Under normal circumstances this should not have any effect from user's point of view, but there is a theoretical chance of the cache being out of sync with the table in case the table is modified directly (e.g. by some C code) without removing attributes.

In cases where the table object cannot be modified (or such modification would not survive) fmatch.hash can be used to build the hash table and return table object including the hash table. In that case no lookup is done and x is only used to determine the type into which table needs to be coerced.

Also fmatch does not convert to a common encoding so strings with different representation in two encodings don't match.


Simon Urbanek

See Also



# some random speed comparison examples:
# first use integer matching
x = as.integer(rnorm(1e6) * 1000000)
s = 1:100
# the first call to fmatch is comparable to match
# but the subsequent calls take no time!
# here is the speed of match for comparison
system.time(base::match(s, x))
# the results should be identical
identical(base::match(s, x), fmatch(s, x))

# next, match a factor against the table
# this will require both x and the factor
# to be cast to strings
s = factor(c("1","1","2","foo","3",NA))
# because the casting will have to allocate a string
# cache in R, we run a dummy conversion to take
# that out of the equation
dummy = as.character(x)
# now we can run the speed tests
system.time(fmatch(s, x))
system.time(fmatch(s, x))
# the cache is still valid for string matches as well
# now back to match
system.time(base::match(s, x))
identical(base::match(s, x), fmatch(s, x))

# finally, some reals to match
y = rnorm(1e6)
s = c(y[sample(length(y), 100)], 123.567, NA, NaN)
system.time(fmatch(s, y))
system.time(fmatch(s, y))
system.time(fmatch(s, y))
system.time(base::match(s, y))
identical(base::match(s, y), fmatch(s, y))

# this used to fail before 0.1-2 since nomatch was ignored
identical(base::match(4L, 1:3, nomatch=0), fmatch(4L, 1:3, nomatch=0))

fastmatch documentation built on Aug. 18, 2023, 9:07 a.m.

Related to fmatch in fastmatch...