I've been interested in optimizing "renumbering" algorithms that can relabel an arbitrary array of integers with duplicates into labels starting from 1. Sets and maps are too slow for what I've been trying to do, as are sorts. Is there a data structure that only remembers if a number has been seen or not reliably? I was considering experimenting with a bloom filter, but I have >12M integers and the target performance is faster than a good hashmap. Is this possible?
Here's a simple example pseudo-c++ algorithm that would be slow:
// note: all elements guaranteed > 0
std::vector<uint64_t> arr = { 21942198, 91292, 21942198, ... millions more };
std::unordered_map<uint64_t, uint64_t> renumber;
renumber.reserve(arr.size());
uint64_t next_label = 1;
for (uint64_t i = 0; i < arr.size(); i++) {
uint64_t elem = arr[i];
if (renumber[elem]) {
arr[i] = renumber[elem];
}
else {
renumber[elem] = next_label;
arr[i] = next_label;
++next_label;
}
}
Example input/output:
{ 12, 38, 1201, 120, 12, 39, 320, 1, 1 }
->
{ 1, 2, 3, 4, 1, 5, 6, 7, 7 }
Your algorithm is not bad, but the appropriate data structure to use for the map is a hash table with open addressing.
As explained in this answer, std::unordered_map can't be implemented that way: https://stackoverflow.com/a/31113618/5483526
So if the STL container is really too slow for you, then you can do better by making your own.
Note, however, that:
90% of the time, when someone complains about STL containers being too slow, they are running a debug build with optimizations turned off. Make sure you are running a release build compiled with optimizations on. Running your code on 12M integers should take a few milliseconds at most.
You are accessing the map multiple times when only once is required, like this:
uint64_t next_label = 1;
for (size_t i = 0; i < arr.size(); i++) {
uint64_t elem = arr[i];
uint64_t &label = renumber[elem];
if (!label) {
label = next_label++;
}
arr[i] = label;
}
Note that the unordered_map operator [] returns a reference to the associated value (creating it if it doesn't exist), so you can test and modify the value without having to search the map again.
Updated with bug fix
First, anytime you experience "slowness" with a std:: collection class like vector or map, just recompile with optimizations (release build). There is usually a 10x speedup.
Now to your problem. I'll show a two-pass solution that runs in O(N) time. I'll leave it as an exercise for you to convert to a one-pass solution. But I'll assert that this should be fast enough, even for vectors with millions of items.
First, declare not one, but two unordered maps:
std::unordered_map<uint64_t, uint64_t> element_to_label;
std::unordered_map<uint64_t, std::pair<uint64_t, std::vector<uint64_t>>> label_to_elements;
The first map, element_to_label maps an integer value found in the original array to it's unique label.
The second map, label_to_elements maps to both the element value and the list of indices that element occurs in the original array.
Now to build these maps:
element_to_label.reserve(arr.size());
label_to_elements.reserve(arr.size());
uint64_t next_label = 1;
for (size_t index = 0; index < arr.size(); index++)
{
const uint64_t elem = arr[index];
auto itor = element_to_label.find(elem);
if (itor == element_to_label.end())
{
// new element
element_to_label[elem] = next_label;
auto &p = label_to_elements[next_label];
p.first = elem;
p.second.push_back(index);
next_label++;
}
else
{
// existing element
uint64_t label = itor->second;
label_to_elements[label].second.push_back(index);
}
}
When the above code runs, it's built up a database all values in the array, their labels, and indices where they occur.
So now to renumber the array such that all elements are replaced with their smaller label value:
for (auto itor = label_to_elements.begin(); itor != label_to_elements.end(); itor++)
{
uint64_t label = itor->first;
auto& p = itor->second;
uint64_t elem = p.first; // technically, this isn't needed. It's just useful to know which element value we are replacing from the original array
const auto& vec = p.second;
for (size_t j = 0; j < vec.size(); j++)
{
size_t index = vec[j];
arr[index] = label;
}
}
Notice where I assign variables by reference with the & operator to avoid making an expensive copy of any value in the maps.
So if your original vector or array was this:
{ 100001, 2000002, 300003, 400004, 400004, 300003, 2000002, 100001 };
Then the application of labels would render the array as this:
{1,2,3,4,4,3,2,1}
And what's nice you still have a quick O(1) look operator to map any label in that set back to its original element value using label_to_elements
c++ newbie here. So for an assignment I have to rotate all the elements in a vector to the left one. So, for instance, the elements {1,2,3} should rotate to {2,3,1}.
I'm researching how to do it, and I saw the rotate() function, but I don't think that will work given my code. And then I saw a for loop that could do it, but I'm not sure how to translate that into a return statement. (i tried to adjust it and failed)
This is what I have so far, but it is very wrong (i haven't gotten a single result that hasn't ended in an error yet)
Edit: The vector size I have to deal with is just three, so it doesn't need to account for any sized vector
#include <vector>
using namespace std;
vector<int> rotate(const vector<int>& v)
{
// PUT CODE BELOW THIS LINE. DON'T CHANGE ANYTHING ABOVE.
vector<int> result;
int size = 3;
for (auto i = 0; i < size - 1; ++i)
{
v.at(i) = v.at(i + 1);
result.at(i) = v.at(i);
}
return result;
// PUT CODE ABOVE THIS LINE. DON'T CHANGE ANYTHING BELOW.
}
All my teacher does it upload textbook pages that explain what certain parts of code are supposed to do but the textbook pages offer NO help in trying to figure out how to actually apply this stuff.
So could someone please give me a few pointers?
Since you know exactly how many elements you have, and it's the smallest number that makes sense to rotate, you don't need to do anything fancy - just place the items in the order that you need, and return the result:
vector<int> rotate3(const vector<int>& x) {
return vector<int> { x[1], x[2], x[0] };
}
Note that if your collection always has three elements, you could use std::array instead:
std::array<int,3>
First, just pay attention that you have passed v as const reference (const vector<int>&) so you are forbbiden to modify the state of v in v.at(i) = v.at(i + 1);
Although Sergey has already answered a straight forward solution, you could correct your code like this:
#include <vector>
using namespace std;
vector<int> left_rotate(const vector<int>& v)
{
vector<int> result;
int size = v.size(); // this way you are able to rotate vectors of any size
for (auto i = 1; i < size; ++i)
result.push_back(v.at(i));
// adding first element of v at end of result
result.push_back(v.front());
return result;
}
Use Sergey's answer. This answer deals with why what the asker attempted did not work. They're damn close, so it's worth going though it, explaining the problems, and showing how to fix it.
In
v.at(i) = v.at(i + 1);
v is constant. You can't write to it. The naïve solution (which won't work) is to cut out the middle-man and write directly to the result vector because it is NOT const
result.at(i) = v.at(i + 1);
This doesn't work because
vector<int> result;
defines an empty vector. There is no at(i) to write to, so at throws an exception that terminates the program.
As an aside, the [] operator does not check bounds like at does and will not throw an exception. This can lead you to thinking the program worked when instead it was writing to memory the vector did not own. This would probably crash the program, but it doesn't have to1.
The quick fix here is to ensure usable storage with
vector<int> result(v.size());
The resulting code
vector<int> rotate(const vector<int>& v)
{
// PUT CODE BELOW THIS LINE. DON'T CHANGE ANYTHING ABOVE.
vector<int> result(v.size()); // change here to size the vector
int size = 3;
for (auto i = 0; i < size - 1; ++i)
{
result.at(i) = v.at(i + 1); // change here to directly assign to result
}
return result;
// PUT CODE ABOVE THIS LINE. DON'T CHANGE ANYTHING BELOW.
}
almost works. But when we run it on {1, 2, 3} result holds {2, 3, 0} at the end. We lost the 1. That's because v.at(i + 1) never touches the first element of v. We could increase the number of for loop iterations and use the modulo operator
vector<int> rotate(const vector<int>& v)
{
// PUT CODE BELOW THIS LINE. DON'T CHANGE ANYTHING ABOVE.
vector<int> result(v.size());
int size = 3;
for (auto i = 0; i < size; ++i) // change here to iterate size times
{
result.at(i) = v.at((i + 1) % size); // change here to make i + 1 wrap
}
return result;
// PUT CODE ABOVE THIS LINE. DON'T CHANGE ANYTHING BELOW.
}
and now the output is {2, 3, 1}. But it's just as easy, and probably a bit faster, to just do what we were doing and tack on the missing element after the loop.
vector<int> rotate(const vector<int>& v)
{
// PUT CODE BELOW THIS LINE. DON'T CHANGE ANYTHING ABOVE.
vector<int> result(v.size());
int size = 3;
for (auto i = 0; i < size - 1; ++i)
{
result.at(i) = v.at(i + 1);
}
result.at(size - 1) = v.at(0); // change here to store first element
return result;
// PUT CODE ABOVE THIS LINE. DON'T CHANGE ANYTHING BELOW.
}
Taking this a step further, the size of three is an unnecessary limitation for this function that I would get rid of and since we're guaranteeing that we never go out of bounds in our for loop, we don't need the extra testing in at
vector<int> rotate(const vector<int>& v)
{
// PUT CODE BELOW THIS LINE. DON'T CHANGE ANYTHING ABOVE.
if (v.empty()) // nothing to rotate.
{
return vector<int>{}; // return empty result
}
vector<int> result(v.size());
for (size_t i = 0; i < v.size() - 1; ++i) // Explicitly using size_t because
// 0 is an int, and v.size() is an
// unsigned integer of implementation-
// defined size but cannot be larger
// than size_t
// note v.size() - 1 is only safe because
// we made sure v is not empty above
// otherwise 0 - 1 in unsigned math
// Becomes a very, very large positive
// number
{
result[i] = v[i + 1];
}
result.back() = v.front(); // using direct calls to front and back because it's
// a little easier on my mind than math and []
return result;
// PUT CODE ABOVE THIS LINE. DON'T CHANGE ANYTHING BELOW.
}
We can go further still and use iterators and range-based for loops, but I think this is enough for now. Besides at the end of the day, you throw the function out completely and use std::rotate from the <algorithm> library.
1This is called Undefined Behaviour (UB), and one of the most fearsome things about UB is anything could happen including giving you the expected result. We put up with UB because it makes for very fast, versatile programs. Validity checks are not made where you don't need them (along with where you did) unless the compiler and library writers decide to make those checks and give guaranteed behaviour like an error message and crash. Microsoft, for example, does exactly this in the vector implementation in the implementation used when you make a debug build. The release version of Microsoft's vector make no checks and assumes you wrote the code correctly and would prefer the executable to be as fast as possible.
I saw the rotate() function, but I don't think that will work given my code.
Yes it will work.
When learning there is gain in "reinventing the wheel" (e.g. implementing rotate yourself) and there is also gain in learning how to use the existing pieces (e.g. use standard library algorithm functions).
Here is how you would use std::rotate from the standard library:
std::vector<int> rotate_1(const std::vector<int>& v)
{
std::vector<int> result = v;
std::rotate(result.begin(), result.begin() + 1, result.end());
return result;
}
I'm working with a huge amount of data stored in an array, and am trying to optimize the amount of time it takes to access and modify it. I'm using Window, c++ and VS2015 (Release mode).
I ran some tests and don't really understand the results I'm getting, so I would love some help optimizing my code.
First, let's say I have the following class:
class foo
{
public:
int x;
foo()
{
x = 0;
}
void inc()
{
x++;
}
int X()
{
return x;
}
void addX(int &_x)
{
_x++;
}
};
I start by initializing 10 million pointers to instances of that class into a std::vector of the same size.
#include <vector>
int count = 10000000;
std::vector<foo*> fooArr;
fooArr.resize(count);
for (int i = 0; i < count; i++)
{
fooArr[i] = new foo();
}
When I run the following code, and profile the amount of time it takes to complete, it takes approximately 350ms (which, for my purposes, is far too slow):
for (int i = 0; i < count; i++)
{
fooArr[i]->inc(); //increment all elements
}
To test how long it takes to increment an integer that many times, I tried:
int x = 0;
for (int i = 0; i < count; i++)
{
x++;
}
Which returns in <1ms.
I thought maybe the number of integers being changed was the problem, but the following code still takes 250ms, so I don't think it's that:
for (int i = 0; i < count; i++)
{
fooArr[0]->inc(); //only increment first element
}
I thought maybe the array index access itself was the bottleneck, but the following code takes <1ms to complete:
int x;
for (int i = 0; i < count; i++)
{
x = fooArr[i]->X(); //set x
}
I thought maybe the compiler was doing some hidden optimizations on the loop itself for the last example (since the value of x will be the same during each iteration of the loop, so maybe the compiler skips unnecessary iterations?). So I tried the following, and it takes 350ms to complete:
int x;
for (int i = 0; i < count; i++)
{
fooArr[i]->addX(x); //increment x inside foo function
}
So that one was slow again, but maybe only because I'm incrementing an integer with a pointer again.
I tried the following too, and it returns in 350ms as well:
for (int i = 0; i < count; i++)
{
fooArr[i]->x++;
}
So am I stuck here? Is ~350ms the absolute fastest that I can increment an integer, inside of 10million pointers in a vector? Or am I missing some obvious thing? I experimented with multithreading (giving each thread a different chunk of the array to increment) and that actually took longer once I started using enough threads. Maybe that was due to some other obvious thing I'm missing, so for now I'd like to stay away from multithreading to keep things simple.
I'm open to trying containers other than a vector too, if it speeds things up, but whatever container I end up using, I need to be able to easily resize it, remove elements, etc.
I'm fairly new to c++ so any help would be appreciated!
Let's look from the CPU point of view.
Incrementing an integer means I have it in a CPU register and just increments it. This is the fastest option.
I'm given an address (vector->member) and I must copy it to a register, increment, and copy the result back to the address. Worst: My CPU cache is filled with vector pointers, not with vector-member pointers. Too few hits, too much cache "refueling".
If I could manage to have all those members just in a vector, CPU cache hits would be much more frequent.
Try the following:
int count = 10000000;
std::vector<foo> fooArr;
fooArr.resize(count, foo());
for (auto it= fooArr.begin(); it != fooArr.end(); ++it) {
it->inc();
}
The new is killing you and actually you don't need it because resize inserts elements at the end if the size it's greater (check the docs: std::vector::resize)
And the other thing it's about using pointers which IMHO should be avoided until the last moment and it's uneccesary in this case. The performance should be a little bit faster in this case since you get better locality of your references (see cache locality). If they were polymorphic or something more complicated it might be different.
I have gathered a large amount of extremely useful information from other peoples' questions and answers on SO, and have searched duly for an answer to this one as well. Unfortunately I have not found a solution to this problem.
The following function to generate a list of primes:
void genPrimes (std::vector<int>* primesPtr, int upperBound = 10)
{
std::ofstream log;
log.open("log.txt");
std::vector<int>& primesRef = *primesPtr;
// Populate primes with non-neg reals
for (int i = 2; i <= upperBound; i++)
primesRef.push_back(i);
log << "Generated reals successfully." << std::endl;
log << primesRef.size() << std::endl;
// Eratosthenes sieve to remove non-primes
for (int i = 0; i < primesRef.size(); i++) {
if (primesRef[i] == 0) continue;
int jumpStart = primesRef[i];
for (int jump = jumpStart; jump < primesRef.size(); jump += jumpStart) {
if (primesRef[i+jump] == 0) continue;
primesRef[i+jump] = 0;
}
}
log << "Executed Eratosthenes Sieve successfully.\n";
for (int i = 0; i < primesRef.size(); i++) {
if (primesRef[i] == 0) {
primesRef.erase(primesRef.begin() + i);
i--;
}
}
log << "Cleaned list.\n";
log.close();
}
is called by:
const int SIZE = 500;
std::vector<int>* primes = new std::vector<int>[SIZE];
genPrimes(primes, SIZE);
This code works well. However, when I change the value of SIZE to a larger number (say, 500000), the compiler returns a "segmentation error." I'm not familiar enough with vectors to understand the problem. Any help is much appreciated.
You are accessing primesRef[i + jump] where i could be primesRef.size() - 1 and jump could be primesRef.size() - 1, leading to an out of bounds access.
It is happening with a 500 limit, it is just that you happen to not have any bad side effects from the out of bound access at the moment.
Also note that using a vector here is a bad choice as every erase will have to move all of the following entries in memory.
Are you sure you wanted to do
new std::vector<int> [500];
and not
new std::vector<int> (500);
In the latter case, you are specifying the size of the vector, whose location is available to you via the variable named 'primes'.
In the former, you are requesting space for 500 vectors, each sized to the default that the STL library wants.
That would be something like (on my system : 24*500 bytes). In the latter case, 500 length vector(only one vector) is what you are asking for.
EDIT: look at the usage - he needs just one vector.
std::vector& primesRef = *primesPtr;
The problem lies here:
// Populate primes with non-neg reals
for (int i = 2; i <= upperBound; i++)
primesRef.push_back(i);
You only have N-2 elements in your vector pushed back, but then try to access an element at N-1 (i+jump). The fact that it did not fail on 500 is just dumb luck that the memory being overwritten was not catastrophic.
This code works well. However, when I change the value of SIZE to a larger number (say, 500000), ...
That may blow your stack, and be to big allocated with it. You need dynamic memory allocation for all of the std::vector<int> instances you believe to need.
To achieve that, simply use a nested std::vetcor like this.
std::vector<std::vector<int>> primes(SIZE);
instead.
But to get straight on, I seriously doubt you need number of SIZE vector instances to store all of the prime numbers found, but just a single one initialized like this:
std::vector<int> primes(SIZE);
This question already has answers here:
How do I sort a std::vector by the values of a different std::vector? [duplicate]
(13 answers)
Closed 12 months ago.
I'd like to reorder the items in a vector, using another vector to specify the order:
char A[] = { 'a', 'b', 'c' };
size_t ORDER[] = { 1, 0, 2 };
vector<char> vA(A, A + sizeof(A) / sizeof(*A));
vector<size_t> vOrder(ORDER, ORDER + sizeof(ORDER) / sizeof(*ORDER));
reorder_naive(vA, vOrder);
// A is now { 'b', 'a', 'c' }
The following is an inefficient implementation that requires copying the vector:
void reorder_naive(vector<char>& vA, const vector<size_t>& vOrder)
{
assert(vA.size() == vOrder.size());
vector vCopy = vA; // Can we avoid this?
for(int i = 0; i < vOrder.size(); ++i)
vA[i] = vCopy[ vOrder[i] ];
}
Is there a more efficient way, for example, that uses swap()?
This algorithm is based on chmike's, but the vector of reorder indices is const. This function agrees with his for all 11! permutations of [0..10]. The complexity is O(N^2), taking N as the size of the input, or more precisely, the size of the largest orbit.
See below for an optimized O(N) solution which modifies the input.
template< class T >
void reorder(vector<T> &v, vector<size_t> const &order ) {
for ( int s = 1, d; s < order.size(); ++ s ) {
for ( d = order[s]; d < s; d = order[d] ) ;
if ( d == s ) while ( d = order[d], d != s ) swap( v[s], v[d] );
}
}
Here's an STL style version which I put a bit more effort into. It's about 47% faster (that is, almost twice as fast over [0..10]!) because it does all the swaps as early as possible and then returns. The reorder vector consists of a number of orbits, and each orbit is reordered upon reaching its first member. It's faster when the last few elements do not contain an orbit.
template< typename order_iterator, typename value_iterator >
void reorder( order_iterator order_begin, order_iterator order_end, value_iterator v ) {
typedef typename std::iterator_traits< value_iterator >::value_type value_t;
typedef typename std::iterator_traits< order_iterator >::value_type index_t;
typedef typename std::iterator_traits< order_iterator >::difference_type diff_t;
diff_t remaining = order_end - 1 - order_begin;
for ( index_t s = index_t(), d; remaining > 0; ++ s ) {
for ( d = order_begin[s]; d > s; d = order_begin[d] ) ;
if ( d == s ) {
-- remaining;
value_t temp = v[s];
while ( d = order_begin[d], d != s ) {
swap( temp, v[d] );
-- remaining;
}
v[s] = temp;
}
}
}
And finally, just to answer the question once and for all, a variant which does destroy the reorder vector (filling it with -1's). For permutations of [0..10], It's about 16% faster than the preceding version. Because overwriting the input enables dynamic programming, it is O(N), asymptotically faster for some cases with longer sequences.
template< typename order_iterator, typename value_iterator >
void reorder_destructive( order_iterator order_begin, order_iterator order_end, value_iterator v ) {
typedef typename std::iterator_traits< value_iterator >::value_type value_t;
typedef typename std::iterator_traits< order_iterator >::value_type index_t;
typedef typename std::iterator_traits< order_iterator >::difference_type diff_t;
diff_t remaining = order_end - 1 - order_begin;
for ( index_t s = index_t(); remaining > 0; ++ s ) {
index_t d = order_begin[s];
if ( d == (diff_t) -1 ) continue;
-- remaining;
value_t temp = v[s];
for ( index_t d2; d != s; d = d2 ) {
swap( temp, v[d] );
swap( order_begin[d], d2 = (diff_t) -1 );
-- remaining;
}
v[s] = temp;
}
}
In-place reordering of vector
Warning: there is an ambiguity about the semantic what the ordering-indices mean. Both are answered here
move elements of vector to the position of the indices
Interactive version here.
#include <iostream>
#include <vector>
#include <assert.h>
using namespace std;
void REORDER(vector<double>& vA, vector<size_t>& vOrder)
{
assert(vA.size() == vOrder.size());
// for all elements to put in place
for( int i = 0; i < vA.size() - 1; ++i )
{
// while the element i is not yet in place
while( i != vOrder[i] )
{
// swap it with the element at its final place
int alt = vOrder[i];
swap( vA[i], vA[alt] );
swap( vOrder[i], vOrder[alt] );
}
}
}
int main()
{
std::vector<double> vec {7, 5, 9, 6};
std::vector<size_t> inds {1, 3, 0, 2};
REORDER(vec, inds);
for (size_t vv = 0; vv < vec.size(); ++vv)
{
std::cout << vec[vv] << std::endl;
}
return 0;
}
output
9
7
6
5
note that you can save one test because if n-1 elements are in place the last nth element is certainly in place.
On exit vA and vOrder are properly ordered.
This algorithm performs at most n-1 swapping because each swap moves the element to its final position. And we'll have to do at most 2N tests on vOrder.
draw the elements of vector from the position of the indices
Try it interactively here.
#include <iostream>
#include <vector>
#include <assert.h>
template<typename T>
void reorder(std::vector<T>& vec, std::vector<size_t> vOrder)
{
assert(vec.size() == vOrder.size());
for( size_t vv = 0; vv < vec.size() - 1; ++vv )
{
if (vOrder[vv] == vv)
{
continue;
}
size_t oo;
for(oo = vv + 1; oo < vOrder.size(); ++oo)
{
if (vOrder[oo] == vv)
{
break;
}
}
std::swap( vec[vv], vec[vOrder[vv]] );
std::swap( vOrder[vv], vOrder[oo] );
}
}
int main()
{
std::vector<double> vec {7, 5, 9, 6};
std::vector<size_t> inds {1, 3, 0, 2};
reorder(vec, inds);
for (size_t vv = 0; vv < vec.size(); ++vv)
{
std::cout << vec[vv] << std::endl;
}
return 0;
}
Output
5
6
7
9
It appears to me that vOrder contains a set of indexes in the desired order (for example the output of sorting by index). The code example here follows the "cycles" in vOrder, where following a sub-set (could be all of vOrder) of indexes will cycle through the sub-set, ending back at the first index of the sub-set.
Wiki article on "cycles"
https://en.wikipedia.org/wiki/Cyclic_permutation
In the following example, every swap places at least one element in it's proper place. This code example effectively reorders vA according to vOrder, while "unordering" or "unpermuting" vOrder back to its original state (0 :: n-1). If vA contained the values 0 through n-1 in order, then after reorder, vA would end up where vOrder started.
template <class T>
void reorder(vector<T>& vA, vector<size_t>& vOrder)
{
assert(vA.size() == vOrder.size());
// for all elements to put in place
for( size_t i = 0; i < vA.size(); ++i )
{
// while vOrder[i] is not yet in place
// every swap places at least one element in it's proper place
while( vOrder[i] != vOrder[vOrder[i]] )
{
swap( vA[vOrder[i]], vA[vOrder[vOrder[i]]] );
swap( vOrder[i], vOrder[vOrder[i]] );
}
}
}
This can also be implemented a bit more efficiently using moves instead swaps. A temp object is needed to hold an element during the moves. Example C code, reorders A[] according to indexes in I[], also sorts I[] :
void reorder(int *A, int *I, int n)
{
int i, j, k;
int tA;
/* reorder A according to I */
/* every move puts an element into place */
/* time complexity is O(n) */
for(i = 0; i < n; i++){
if(i != I[i]){
tA = A[i];
j = i;
while(i != (k = I[j])){
A[j] = A[k];
I[j] = j;
j = k;
}
A[j] = tA;
I[j] = j;
}
}
}
If it is ok to modify the ORDER array then an implementation that sorts the ORDER vector and at each sorting operation also swaps the corresponding values vector elements could do the trick, I think.
A survey of existing answers
You ask if there is "a more efficient way". But what do you mean by efficient and what are your requirements?
Potatoswatter's answer works in O(N²) time with O(1) additional space and doesn't mutate the reordering vector.
chmike and rcgldr give answers which use O(N) time with O(1) additional space, but they achieve this by mutating the reordering vector.
Your original answer allocates new space and then copies data into it while Tim MB suggests using move semantics. However, moving still requires a place to move things to and an object like an std::string has both a length variable and a pointer. In other words, a move-based solution requires O(N) allocations for any objects and O(1) allocations for the new vector itself. I explain why this is important below.
Preserving the reordering vector
We might want that reordering vector! Sorting costs O(N log N). But, if you know you'll be sorting several vectors in the same way, such as in a Structure of Arrays (SoA) context, you can sort once and then reuse the results. This can save a lot of time.
You might also want to sort and then unsort data. Having the reordering vector allows you to do this. A use case here is for performing genomic sequencing on GPUs where maximal speed efficiency is obtained by having sequences of similar lengths processed in batches. We cannot rely on the user providing sequences in this order so we sort and then unsort.
So, what if we want the best of all worlds: O(N) processing without the costs of additional allocation but also without mutating our ordering vector (which we might, after all, want to reuse)? To find that world, we need to ask:
Why is extra space bad?
There are two reasons you might not want to allocate additional space.
The first is that you don't have much space to work with. This can occur in two situations: you're on an embedded device with limited memory. Usually this means you're working with small datasets, so the O(N²) solution is probably fine here. But it can also happen when you are working with really large datasets. In this case O(N²) is unacceptable and you have to use one of the O(N) mutating solutions.
The other reason extra space is bad is because allocation is expensive. For smaller datasets it can cost more than the actual computation. Thus, one way to achieve efficiency is to eliminate allocation.
Outline
When we mutate the ordering vector we are doing so as a way to indicate whether elements are in their permuted positions. Rather than doing this, we could use a bit-vector to indicate that same information. However, if we allocate the bit vector each time that would be expensive.
Instead, we could clear the bit vector each time by resetting it to zero. However, that incurs an additional O(N) cost per function use.
Rather, we can store a "version" value in a vector and increment this on each function use. This gives us O(1) access, O(1) clear, and an amoritzed allocation cost. This works similarly to a persistent data structure. The downside is that if we use an ordering function too often the version counter needs to be reset, though the O(N) cost of doing so is amortized.
This raises the question: what is the optimal data type for the version vector? A bit-vector maximizes cache utilization but requires a full O(N) reset after each use. A 64-bit data type probably never needs to be reset, but has poor cache utilization. Experimenting is the best way to figure this out.
Two types of permutations
We can view an ordering vector as having two senses: forward and backward. In the forward sense, the vector tell us where elements go to. In the backward sense, the vector tells us where elements are coming from. Since the ordering vector is implicitly a linked list, the backward sense requires O(N) additional space, but, again, we can amortize the allocation cost. Applying the two senses sequentially brings us back to our original ordering.
Performance
Running single-threaded on my "Intel(R) Xeon(R) E-2176M CPU # 2.70GHz", the following code takes about 0.81ms per reordering for sequences 32,767 elements long.
Code
Fully commented code for both senses with tests:
#include <algorithm>
#include <cassert>
#include <random>
#include <stack>
#include <stdexcept>
#include <vector>
///#brief Reorder a vector by moving its elements to indices indicted by another
/// vector. Takes O(N) time and O(N) space. Allocations are amoritzed.
///
///#param[in,out] values Vector to be reordered
///#param[in] ordering A permutation of the vector
///#param[in,out] visited A black-box vector to be reused between calls and
/// shared with with `backward_reorder()`
template<class ValueType, class OrderingType, class ProgressType>
void forward_reorder(
std::vector<ValueType> &values,
const std::vector<OrderingType> &ordering,
std::vector<ProgressType> &visited
){
if(ordering.size()!=values.size()){
throw std::runtime_error("ordering and values must be the same size!");
}
//Size the visited vector appropriately. Since vectors don't shrink, this will
//shortly become large enough to handle most of the inputs. The vector is 1
//larger than necessary because the first element is special.
if(visited.empty() || visited.size()-1<values.size());
visited.resize(values.size()+1);
//If the visitation indicator becomes too large, we reset everything. This is
//O(N) expensive, but unlikely to occur in most use cases if an appropriate
//data type is chosen for the visited vector. For instance, an unsigned 32-bit
//integer provides ~4B uses before it needs to be reset. We subtract one below
//to avoid having to think too much about off-by-one errors. Note that
//choosing the biggest data type possible is not necessarily a good idea!
//Smaller data types will have better cache utilization.
if(visited.at(0)==std::numeric_limits<ProgressType>::max()-1)
std::fill(visited.begin(), visited.end(), 0);
//We increment the stored visited indicator and make a note of the result. Any
//value in the visited vector less than `visited_indicator` has not been
//visited.
const auto visited_indicator = ++visited.at(0);
//For doing an early exit if we get everything in place
auto remaining = values.size();
//For all elements that need to be placed
for(size_t s=0;s<ordering.size() && remaining>0;s++){
assert(visited[s+1]<=visited_indicator);
//Ignore already-visited elements
if(visited[s+1]==visited_indicator)
continue;
//Don't rearrange if we don't have to
if(s==visited[s])
continue;
//Follow this cycle, putting elements in their places until we get back
//around. Use move semantics for speed.
auto temp = std::move(values[s]);
auto i = s;
for(;s!=(size_t)ordering[i];i=ordering[i],--remaining){
std::swap(temp, values[ordering[i]]);
visited[i+1] = visited_indicator;
}
std::swap(temp, values[s]);
visited[i+1] = visited_indicator;
}
}
///#brief Reorder a vector by moving its elements to indices indicted by another
/// vector. Takes O(2N) time and O(2N) space. Allocations are amoritzed.
///
///#param[in,out] values Vector to be reordered
///#param[in] ordering A permutation of the vector
///#param[in,out] visited A black-box vector to be reused between calls and
/// shared with with `forward_reorder()`
template<class ValueType, class OrderingType, class ProgressType>
void backward_reorder(
std::vector<ValueType> &values,
const std::vector<OrderingType> &ordering,
std::vector<ProgressType> &visited
){
//The orderings form a linked list. We need O(N) memory to reverse a linked
//list. We use `thread_local` so that the function is reentrant.
thread_local std::stack<OrderingType> stack;
if(ordering.size()!=values.size()){
throw std::runtime_error("ordering and values must be the same size!");
}
//Size the visited vector appropriately. Since vectors don't shrink, this will
//shortly become large enough to handle most of the inputs. The vector is 1
//larger than necessary because the first element is special.
if(visited.empty() || visited.size()-1<values.size());
visited.resize(values.size()+1);
//If the visitation indicator becomes too large, we reset everything. This is
//O(N) expensive, but unlikely to occur in most use cases if an appropriate
//data type is chosen for the visited vector. For instance, an unsigned 32-bit
//integer provides ~4B uses before it needs to be reset. We subtract one below
//to avoid having to think too much about off-by-one errors. Note that
//choosing the biggest data type possible is not necessarily a good idea!
//Smaller data types will have better cache utilization.
if(visited.at(0)==std::numeric_limits<ProgressType>::max()-1)
std::fill(visited.begin(), visited.end(), 0);
//We increment the stored visited indicator and make a note of the result. Any
//value in the visited vector less than `visited_indicator` has not been
//visited.
const auto visited_indicator = ++visited.at(0);
//For doing an early exit if we get everything in place
auto remaining = values.size();
//For all elements that need to be placed
for(size_t s=0;s<ordering.size() && remaining>0;s++){
assert(visited[s+1]<=visited_indicator);
//Ignore already-visited elements
if(visited[s+1]==visited_indicator)
continue;
//Don't rearrange if we don't have to
if(s==visited[s])
continue;
//The orderings form a linked list. We need to follow that list to its end
//in order to reverse it.
stack.emplace(s);
for(auto i=s;s!=(size_t)ordering[i];i=ordering[i]){
stack.emplace(ordering[i]);
}
//Now we follow the linked list in reverse to its beginning, putting
//elements in their places. Use move semantics for speed.
auto temp = std::move(values[s]);
while(!stack.empty()){
std::swap(temp, values[stack.top()]);
visited[stack.top()+1] = visited_indicator;
stack.pop();
--remaining;
}
visited[s+1] = visited_indicator;
}
}
int main(){
std::mt19937 gen;
std::uniform_int_distribution<short> value_dist(0,std::numeric_limits<short>::max());
std::uniform_int_distribution<short> len_dist (0,std::numeric_limits<short>::max());
std::vector<short> data;
std::vector<short> ordering;
std::vector<short> original;
std::vector<size_t> progress;
for(int i=0;i<1000;i++){
const int len = len_dist(gen);
data.clear();
ordering.clear();
for(int i=0;i<len;i++){
data.push_back(value_dist(gen));
ordering.push_back(i);
}
original = data;
std::shuffle(ordering.begin(), ordering.end(), gen);
forward_reorder(data, ordering, progress);
assert(original!=data);
backward_reorder(data, ordering, progress);
assert(original==data);
}
}
Never prematurely optimize. Meassure and then determine where you need to optimize and what. You can end with complex code that is hard to maintain and bug-prone in many places where performance is not an issue.
With that being said, do not early pessimize. Without changing the code you can remove half of your copies:
template <typename T>
void reorder( std::vector<T> & data, std::vector<std::size_t> const & order )
{
std::vector<T> tmp; // create an empty vector
tmp.reserve( data.size() ); // ensure memory and avoid moves in the vector
for ( std::size_t i = 0; i < order.size(); ++i ) {
tmp.push_back( data[order[i]] );
}
data.swap( tmp ); // swap vector contents
}
This code creates and empty (big enough) vector in which a single copy is performed in-order. At the end, the ordered and original vectors are swapped. This will reduce the copies, but still requires extra memory.
If you want to perform the moves in-place, a simple algorithm could be:
template <typename T>
void reorder( std::vector<T> & data, std::vector<std::size_t> const & order )
{
for ( std::size_t i = 0; i < order.size(); ++i ) {
std::size_t original = order[i];
while ( i < original ) {
original = order[original];
}
std::swap( data[i], data[original] );
}
}
This code should be checked and debugged. In plain words the algorithm in each step positions the element at the i-th position. First we determine where the original element for that position is now placed in the data vector. If the original position has already been touched by the algorithm (it is before the i-th position) then the original element was swapped to order[original] position. Then again, that element can already have been moved...
This algorithm is roughly O(N^2) in the number of integer operations and thus is theoretically worse in performance time as compare to the initial O(N) algorithm. But it can compensate if the N^2 swap operations (worst case) cost less than the N copy operations or if you are really constrained by memory footprint.
It's an interesting intellectual exercise to do the reorder with O(1) space requirement but in 99.9% of the cases the simpler answer will perform to your needs:
void permute(vector<T>& values, const vector<size_t>& indices)
{
vector<T> out;
out.reserve(indices.size());
for(size_t index: indices)
{
assert(0 <= index && index < values.size());
out.push_back(std::move(values[index]));
}
values = std::move(out);
}
Beyond memory requirements, the only way I can think of this being slower would be due to the memory of out being in a different cache page than that of values and indices.
You could do it recursively, I guess - something like this (unchecked, but it gives the idea):
// Recursive function
template<typename T>
void REORDER(int oldPosition, vector<T>& vA,
const vector<int>& vecNewOrder, vector<bool>& vecVisited)
{
// Keep a record of the value currently in that position,
// as well as the position we're moving it to.
// But don't move it yet, or we'll overwrite whatever's at the next
// position. Instead, we first move what's at the next position.
// To guard against loops, we look at vecVisited, and set it to true
// once we've visited a position.
T oldVal = vA[oldPosition];
int newPos = vecNewOrder[oldPosition];
if (vecVisited[oldPosition])
{
// We've hit a loop. Set it and return.
vA[newPosition] = oldVal;
return;
}
// Guard against loops:
vecVisited[oldPosition] = true;
// Recursively re-order the next item in the sequence.
REORDER(newPos, vA, vecNewOrder, vecVisited);
// And, after we've set this new value,
vA[newPosition] = oldVal;
}
// The "main" function
template<typename T>
void REORDER(vector<T>& vA, const vector<int>& newOrder)
{
// Initialise vecVisited with false values
vector<bool> vecVisited(vA.size(), false);
for (int x = 0; x < vA.size(); x++)
{
REORDER(x, vA, newOrder, vecVisited);
}
}
Of course, you do have the overhead of vecVisited. Thoughts on this approach, anyone?
To iterate through the vector is O(n) operation. Its sorta hard to beat that.
Your code is broken. You cannot assign to vA and you need to use template parameters.
vector<char> REORDER(const vector<char>& vA, const vector<size_t>& vOrder)
{
assert(vA.size() == vOrder.size());
vector<char> vCopy(vA.size());
for(int i = 0; i < vOrder.size(); ++i)
vCopy[i] = vA[ vOrder[i] ];
return vA;
}
The above is slightly more efficient.
It is not clear by the title and the question if the vector should be ordered with the same steps it takes to order vOrder or if vOrder already contains the indexes of the desired order.
The first interpretation has already a satisfying answer (see chmike and Potatoswatter), I add some thoughts about the latter.
If the creation and/or copy cost of object T is relevant
template <typename T>
void reorder( std::vector<T> & data, std::vector<std::size_t> & order )
{
std::size_t i,j,k;
for(i = 0; i < order.size() - 1; ++i) {
j = order[i];
if(j != i) {
for(k = i + 1; order[k] != i; ++k);
std::swap(order[i],order[k]);
std::swap(data[i],data[j]);
}
}
}
If the creation cost of your object is small and memory is not a concern (see dribeas):
template <typename T>
void reorder( std::vector<T> & data, std::vector<std::size_t> const & order )
{
std::vector<T> tmp; // create an empty vector
tmp.reserve( data.size() ); // ensure memory and avoid moves in the vector
for ( std::size_t i = 0; i < order.size(); ++i ) {
tmp.push_back( data[order[i]] );
}
data.swap( tmp ); // swap vector contents
}
Note that the two pieces of code in dribeas answer do different things.
I was trying to use #Potatoswatter's solution to sort multiple vectors by a third one and got really confused by output from using the above functions on a vector of indices output from Armadillo's sort_index. To switch from a vector output from sort_index (the arma_inds vector below) to one that can be used with #Potatoswatter's solution (new_inds below), you can do the following:
vector<int> new_inds(arma_inds.size());
for (int i = 0; i < new_inds.size(); i++) new_inds[arma_inds[i]] = i;
I came up with this solution which has the space complexity of O(max_val - min_val + 1), but it can be integrated with std::sort and benefits from std::sort's O(n log n) decent time complexity.
std::vector<int32_t> dense_vec = {1, 2, 3};
std::vector<int32_t> order = {1, 0, 2};
int32_t max_val = *std::max_element(dense_vec.begin(), dense_vec.end());
std::vector<int32_t> sparse_vec(max_val + 1);
int32_t i = 0;
for(int32_t j: dense_vec)
{
sparse_vec[j] = order[i];
i++;
}
std::sort(dense_vec.begin(), dense_vec.end(),
[&sparse_vec](int32_t i1, int32_t i2) {return sparse_vec[i1] < sparse_vec[i2];});
The following assumptions made while writing this code:
Vector values start from zero.
Vector does not contain repeated values.
We have enough memory to sacrifice in order to use std::sort
This should avoid copying the vector:
void REORDER(vector<char>& vA, const vector<size_t>& vOrder)
{
assert(vA.size() == vOrder.size());
for(int i = 0; i < vOrder.size(); ++i)
if (i < vOrder[i])
swap(vA[i], vA[vOrder[i]]);
}