permuteIter. These functions return iterators for iterating over combinations and permutations. They have a similar interface to
permuteGeneral and currently only work with standard combinations and permutations. They do not yet work with constraints (This will be the focus of the next release).
Added “High Performance Benchmarks” and “Combinatorial Iterators in RcppAlgos” vignettes
In 2.3.5 and 2.3.6, we mistakingly allowed a
constraintFun to be applied to a logical vector which was crashing R. We have corrected this in 2.4.0.
Changed the data type for sizing the index matrix in
permuteGeneral. Originally, we were using
int and when the output was large enough, it was causing an integer overflow thus causing the index matrix to be unproperly sized. We have sinced changed the data type to the recommended
std::size_t (See Is there a max array length limit in C++?)
comboGroupsSample. These functions deal with partitioning a vector/set into groups of equal size. See Combinations in R by Groups. See the related integer sequences A025035-A025042 at https://oeis.orgOEIS (E.g. https://oeis.org/A025036A025036 for Number of partitions of
(1, 2, ..., 4n) into sets of size 4.)
Added vignettes (First version with vignettes)
Added website via the excellent package
Extended general partitions algorithm to multisets. E.g.
comboGeneral(10, 8, freqs = rep(1:5, 2), constraintFun = "sum", comparisonFun = "==", limitConstraints = 55)
Improved constraint algorithm for the general case.
Added support for complex and raw types for all combinatorial functions.
Improved permutation algorithm for all cases (I.e. no rep, with rep, and with multisets).
Added loop unrolling to prime sieve algorithm for improved efficiency.
2.3.3. These issues were arising from finding the first vector to meet the criteria in PartitionRep/Distinct. We also found further issues in the standard functions when the length of the partition was 2. The algorithm would eventually try and access an element of a vector at index -1. These fixes were confirmed by successful Rdevel CMD check under ‘r-devel-ubsan-clang’ via docker using the advice found here: https://knausb.github.io/2017/06/reproducing-a-clang-ubsan-issue/ and http://dirk.eddelbuettel.com/code/sanitizers.html.
Fixed clang-UBSAN issue in
2.3.2. It was caused by populating a vector of ints with values larger than
2^31 - 1.
Added optimized algorithm to
constraintFun = "sum",
comparisonFun = "==", and the vector passed has a special form. This problem is a special case of the subset sum problem.
push_back member function instead of pre-allocating matrix when constraint is applied in combo/permuteGeneral. This alleviates the need to guess the upper limit and subsequently subset as only elements that meet the constraints are added.
Fixed error in
PollardRho.cpp when number passed had factors close to the limit in the predefined lookup table (i.e.
constexpr int64_t FirstOmittedPrime = 3989)
Fixed clang-UBSAN issue in
2.3.1. It was caused by casting extremely large values to
Corrected handling of small values in
Explicitly casted to double for sqrt to silence Solaris
Corrected handling of
All functions now have parallel capabilites via
RcppParallel for thread safe matrix class.
Major overhaul of primeSieve for large primes.
stdThreadMax for obtaining the number of threads available on a machine
Parallel argument as it was causing unpredictable errors on certain platforms. Research is ongoing to correct this for use in future versions. The development version will retain this feature.
UBSAN error that caused by filling a vector of integers with numbers larger than
2^31 - 1.
Parallel to general and sampling functions for increased gains in efficiency.
Logical class is now preserved in combinatorial functions
Added gmp support to combinatorial functions. Now, one can accurately and quickly work with combinations/permutations of large vectors E.g.:
FUN argument to all combinatorial functions. Allows user to pass custom functions to be applied to combinations/permutations.
UBSANerror identified by two different unit tests. In both situations, the problem was occurring as a result of populating a vector of integers with values from a vector of doubles that contained a nan (Not-a-Number). Most information was obtained from Brian J. Knaus’s blog titled : “Reproducing a clang-UBSAN issue” (https://knausb.github.io/2017/06/reproducing-a-clang-ubsan-issue/)
Changed max value and explicitly casted a few values to
int64_t in PollardRho.cpp for efficiency while still maintaining accuracy from
isPrimeRcpp is roughly 10% faster now).
Updated core permutation algorithm for greater efficiency and generality
Added capability of generating specific combinations/permutations
Changed arguments to
rowCap is now
Fixed bug with
divisorSieve when the lower bound was greater than 1 and less than the
sqrt of the upper bound. In the previous version, the numbers in this range would have duplicated values/counts.
Increased efficiency of
numDivisorSieve by a factor of 2.
Updated unit tests for greater coverage. See the function
package_coverage from the package
Corrected precision limits in documentation from
Changed const type in
double to correct “UndefinedBehaviorSanitizer”
Changed examples in
primeFactorizeSieve to reduce check time
RcppAlgos-package man file.
Added the following functions:
primeFactorize (vectorized pollard rho factorization),
divisorsRcpp (vectorized factoring (complete)),
isPrimeRcpp (vectorized primality testing using Miller-Rabin algorithm), &
primeCount (based on the primecount algorithm by Kim Walisch)
Completely revamped the
primeSieve function. It is now a segmented sieve of Eratosthenes with wheel factorization based on primesieve by Kim Walisch.
divisorsSieve (reason for the major version update to
Made the sieving functions more flexible. They are now able to generate results over a range and can also produce named objects.
All number theoretic functions have been made more efficient. Some make use of the fast integer division library
libdivide by ridiculousfish.
Added unit tests.
Removed unnecessary files.
Fixed bug in primeSieve that occurred when a number with a decimal was passed (e.g. 2.01).
Adjusted accepted lower bound for
Fixed bug when non-unique elements are present with factors.
Fixed segmentation fault error highlighted by valgrind check in version
Fixed bug in constraint functions that occurred when
m = 1 and the constraint limit was equal to the last element in
v. It was returning a 2x1 matrix with the same value twice. It is now correctly returning a 1x1 matrix with the correct value 1 time.
Reorganized source code such that all utility functions for the combinatoric functions are now in their own file. Additionally added header for this file.
All combinatoric functions can now utilize the rowCap argument. Before, rowCap only applied to the combinatorial functions with constraints.
comboGeneral can now find all combinations of multisets.
permuteGeneral can utilize the argument
m when dealing with multisets. Before,
permuteGeneral would simply return all permutations of a multiset. Now you can specify the lengths of the output.
Fixed bug that would occur in two edge cases involving the constraint functions.
Slightly modified formatting for
Updated combination algorithms. They are now more than twice as fast.
Updated constraint functions so that memory access is always within container bounds
Consolidated redundant code for outputting different
Rcpp types (e.g.
CharacterMatrix, etc.) via a templated approach.
Added the function permuteGeneral that is analogous to comboGeneral only instead of combinations, it gives all permutations. It has an additional argument (i.e. ‘freqs’) that is used to generate permutations of multisets.
All combinatoric functions now support factor types.
Corrected minor typo in
Fixed minor error regarding explicitly comparing variables to large numbers that are typed out. Simply adding a decimal along with a zero remedies the situation.
ComboConstraint function by removing unnecessary subsetting.
Corrected the errors with respect to the math functions in
C++. Explicitly overloaded the parameters of these functions by casting them to the