Suppose I have a std::vector (let's call it myVec) of size N. What's the simplest way to construct a new vector consisting of a copy of elements X through Y, where 0 <= X <= Y <= N-1? For example, myVec [100000] through myVec [100999] in a vector of size 150000.
If this cannot be done efficiently with a vector, is there another STL datatype that I should use instead?
vector<T>::const_iterator first = myVec.begin() + 100000;
vector<T>::const_iterator last = myVec.begin() + 101000;
vector<T> newVec(first, last);
It's an O(N) operation to construct the new vector, but there isn't really a better way.
O(Y-X), or we'd say O(Z) where Z=Y-X. –
Spathic vector<T> newVec(myVec.begin() + 100000, myVec.begin() + 101000);? –
Crosson newVec already exists and you want to replace it's contents use newVec.asign(first, last). It has the same semantics as the constructor in the answer. –
Acetone auto instead of vector<T>::const_iterator –
Morton constructs the container with the contents of the range [first, last). –
Tactic auto cbegin instead of begin ? auto first = myVec.cbegin() + 100000; auto last = myVec.cbegin() + 101000; –
Nalepka last element is not included was relevant , hence the [x,y) convention, in fact the answer used vector.begin + ..1000 to get ...999 elements –
Tactic last in a toy program: godbolt.org/z/87MsPqnWT It doesn't do a bounds check. Compiler :clang 14.0.0, c++2a –
Flasher Just use the vector constructor.
std::vector<int> data();
// Load Z elements into data so that Z > Y > X
std::vector<int> sub(&data[100000],&data[101000]);
operator[] returns a reference. It is only at the point where you read or write the reference that it would become an access violation. Since we do neither but instead get the address we have not invoked UB,. –
Diaphoretic X[Y] is defined as *(X + Y) in the standard. And the operators &* placed together cancel each other out (assuming they have not been overridden (which they can't be for pointers)). => &X[Y] => &*(X + Y) => (X + Y). Which is also why 10000[data] is a valid expression for array access. –
Diaphoretic This discussion is pretty old, but the simplest one isn't mentioned yet, with list-initialization:
vector<int> subvector = {big_vector.begin() + 3, big_vector.end() - 2};
It requires c++11 or above.
Example usage:
#include <iostream>
#include <vector>
#include <algorithm>
using namespace std;
int main(){
vector<int> big_vector = {5,12,4,6,7,8,9,9,31,1,1,5,76,78,8};
vector<int> subvector = {big_vector.begin() + 3, big_vector.end() - 2};
cout << "Big vector: ";
for_each(big_vector.begin(), big_vector.end(),[](int number){cout << number << ";";});
cout << endl << "Subvector: ";
for_each(subvector.begin(), subvector.end(),[](int number){cout << number << ";";});
cout << endl;
}
Result:
Big vector: 5;12;4;6;7;8;9;9;31;1;1;5;76;78;8;
Subvector: 6;7;8;9;9;31;1;1;5;76;
.end() returns the iterator to the "past the last" element. –
Haemoid vector<int> subvector = {iter1, iter2}, iter2 is exclusive. This does not change depending on your perspective. You said "the last element is the one 2 positions before .end()", but it's not. The last element(76) is the one 3 positions before .end(). Very tiny stuff but let's be careful :) –
Haemoid These days, we use spans! So you would write:
#include <span>
...
auto start_pos = 100000;
auto length = 1000;
auto span_of_myvec = std::make_span(myvec);
auto my_subspan = span_of_myvec.subspan(start_pos, length);
to get a span of 1000 elements of the same type as myvec's. Or a more terse form:
auto my_subspan = std::make_span(myvec).subspan(1000000, 1000);
(but I don't like this as much, since the meaning of each numeric argument is not entirely clear; and it gets worse if the length and start_pos are of the same order of magnitude.)
Anyway, remember that this is not a copy, it's just a view of the data in the vector, so be careful. If you want an actual copy, you could do:
std::vector<T> new_vec(my_subspan.cbegin(), my_subspan.cend());
Notes:
gsl::span - from the Guidelines Support Library. For more information, see: http://www.modernescpp.com/index.php/c-core-guideline-the-guidelines-support-library and https://github.com/martinmoene/gsl-lite for a concrete implementationstd::vector has a gazillion constructors, it's super-easy to fall into one you didn't intend to use, so be careful.cbegin and cend just for the principle ;) std::cbegin etc even. –
Dis std::vector<T>(input_iterator, input_iterator), in your case foo = std::vector<T>(myVec.begin () + 100000, myVec.begin () + 150000);, see for example here
If both are not going to be modified (no adding/deleting items - modifying existing ones is fine as long as you pay heed to threading issues), you can simply pass around data.begin() + 100000 and data.begin() + 101000, and pretend that they are the begin() and end() of a smaller vector.
Or, since vector storage is guaranteed to be contiguous, you can simply pass around a 1000 item array:
T *arrayOfT = &data[0] + 100000;
size_t arrayOfTLength = 1000;
Both these techniques take constant time, but require that the length of data doesn't increase, triggering a reallocation.
You didn't mention what type std::vector<...> myVec is, but if it's a simple type or struct/class that doesn't include pointers, and you want the best efficiency, then you can do a direct memory copy (which I think will be faster than the other answers provided). Here is a general example for std::vector<type> myVec where type in this case is int:
typedef int type; //choose your custom type/struct/class
int iFirst = 100000; //first index to copy
int iLast = 101000; //last index + 1
int iLen = iLast - iFirst;
std::vector<type> newVec;
newVec.resize(iLen); //pre-allocate the space needed to write the data directly
memcpy(&newVec[0], &myVec[iFirst], iLen*sizeof(type)); //write directly to destination buffer from source buffer
std::vector(myVec.begin () + 100000, myVec.begin () + 150000); , wouldn't the longer-version of this produce into exactly the same assembly? –
Raeraeann std::vector<>(iter, iter) to memmove(), if appropriate (if constructor is trivial, for a suitable definition of trivial). –
Superabundant memcpy. Do a std::copy or a constructor which accepts a range (two iterators), and the compiler and the std.library will conspire to call memcpy when appropriate. –
Navelwort You could just use insert
vector<type> myVec { n_elements };
vector<type> newVec;
newVec.insert(newVec.begin(), myVec.begin() + X, myVec.begin() + Y);
You can use STL copy with O(M) performance when M is the size of the subvector.
copy. To use std::copy you need to build the destination first (which means initializing all the members to the their default value). Then you can copy them over. So its two distinct operations. Why not simply use the vector constructor and initialize the members directly to their final values. See the nice answer that has +50 votes. –
Diaphoretic newvec.reserve(10100 - 10000);. ITs definitely an option and technically it will work. But out of the two which are you going to recommend? –
Diaphoretic Suppose there are two vectors.
vector<int> vect1{1, 2, 3, 4};
vector<int> vect2;
Method 1. Using copy function. copy(first_iterator_index, last_iterator_index, back_inserter()) :- This function takes 3 arguments, firstly, the first iterator of old vector. Secondly, the last iterator of old vector and third is back_inserter function to insert values from back.
// Copying vector by copy function
copy(vect1.begin(), vect1.end(), back_inserter(vect2));
Method 2. By using Assign Function. assign(first_iterator_o, last_iterator_o). This method assigns the same values to new vector as old one. This takes 2 arguments, first iterator to old vector and last iterator to old vector.
//Copying vector by assign function
vect2.assign(vect1.begin(), vect1.end());
The only way to project a collection that is not linear time is to do so lazily, where the resulting "vector" is actually a subtype which delegates to the original collection. For example, Scala's List#subseq method create a sub-sequence in constant time. However, this only works if the collection is immutable and if the underlying language sports garbage collection.
Maybe the array_view/span in the GSL library is a good option.
Here is also a single file implementation: array_view.
Copy elements from one vector to another easily
In this example, I am using a vector of pairs to make it easy to understand
`
vector<pair<int, int> > v(n);
//we want half of elements in vector a and another half in vector b
vector<pair<lli, lli> > a(v.begin(),v.begin()+n/2);
vector<pair<lli, lli> > b(v.begin()+n/2, v.end());
//if v = [(1, 2), (2, 3), (3, 4), (4, 5), (5, 6)]
//then a = [(1, 2), (2, 3)]
//and b = [(3, 4), (4, 5), (5, 6)]
//if v = [(1, 2), (2, 3), (3, 4), (4, 5), (5, 6), (6, 7)]
//then a = [(1, 2), (2, 3), (3, 4)]
//and b = [(4, 5), (5, 6), (6, 7)]
'
As you can see you can easily copy elements from one vector to another, if you want to copy elements from index 10 to 16 for example then we would use
vector<pair<int, int> > a(v.begin()+10, v.begin+16);
and if you want elements from index 10 to some index from end, then in that case
vector<pair<int, int> > a(v.begin()+10, v.end()-5);
hope this helps, just remember in the last case v.end()-5 > v.begin()+10
Yet another option:
Useful for instance when moving between a thrust::device_vector and a thrust::host_vector, where you cannot use the constructor.
std::vector<T> newVector;
newVector.reserve(1000);
std::copy_n(&vec[100000], 1000, std::back_inserter(newVector));
Should also be complexity O(N)
You could combine this with top anwer code
vector<T>::const_iterator first = myVec.begin() + 100000;
vector<T>::const_iterator last = myVec.begin() + 101000;
std::copy(first, last, std::back_inserter(newVector));
vector::assign may be another solution
// note: size1 < src.size() && size2 < src.size()
std::vector<int> sub1(size1), sub2(size2);
sub1.assign(src.begin(), src.begin() + size1);
sub2.assign(src.begin(), src.begin() + size2);
Posting this late just for others..I bet the first coder is done by now. For simple datatypes no copy is needed, just revert to good old C code methods.
std::vector <int> myVec;
int *p;
// Add some data here and set start, then
p=myVec.data()+start;
Then pass the pointer p and a len to anything needing a subvector.
notelen must be!! len < myVec.size()-start
© 2022 - 2024 — McMap. All rights reserved.
assignin std::vector, such asv.assign(s.begin()+M, s.begin()+N);– Haemoid