How to access to the index of a 2D vector by at() function?

Question

I am going to initialize a 2D vector in a member function, where the input arguments are planned to be fed into determined indexes within the function argument. Furthermore, I'm not gonna use [] operator to member access due to safety stuffs. How can I use at() function in order to access to the index of a 2D vector, as below?

vector<vector<double>> weight;

void Connection::setWeight(const double& value, const double& i, const double& j)
{
    // The other scheme except: weight[i][j] = value;
}

@MaxBeikirch because it is trivial to figure this out by looking at some documentation? — juanchopanza, Sep 01 '14 at 19:09
@jaunchopanza: Maybe you don't believe me, but I asked this question up here after some time browsing the web to find it for more than one dimension... But thank you for your clarification upon the potential capabilities of the documentation!!! — User, Sep 01 '14 at 19:14
Vectors never have more than one dimension. You need to think in terms of "chains of custody"; you get the inner vector from the outer vector then perform a new operation on that inner vector. — Lightness Races in Orbit, Sep 01 '14 at 19:16
@matinking A vector of vectors is just a vector. But you could try to figure it out by looking at what `operator[]` does. — juanchopanza, Sep 01 '14 at 19:17
@matinking: Please do not approve edits that obviously do not add anything at all to the post. — Lightness Races in Orbit, Sep 01 '14 at 19:18

Cory Kramer · Accepted Answer · 2020-07-28T11:34:41.137

16

You can use at twice. Also note, if i and j are indexes, they really shouldn't be double.

void Connection::setWeight(const double value, const size_t i, const size_t j)
{
    weight.at(i).at(j) = value;
}

edited Jul 28 '20 at 11:34

answered Sep 01 '14 at 19:04

Cory Kramer

114,268
16
167
218

You mean they should integer, just like the intrinsic nature of the index type, right? – User Sep 01 '14 at 19:07
2

`size_t` is actually `unsigned int`, since you cannot use a negative index. – Cory Kramer Sep 01 '14 at 19:08

score 12 · Answer 2 · edited May 23 '17 at 10:31

12

_{Risking to be downvoted for premature optimization, and because it doesn't really answers the question asked, still I would like to remind about a serious problem in such approach.}

Like any multidimensional array, vector of vector is not contiguous, which makes its performance characteristics terribly bad because of lacking of spatial locality and associated cache misses.

You can trivially solve the problem by wrapping a single-dimensional array with a convenient interface of two-dimensional array.

template <class T>
class MyArray2D
{
    std::vector<T> data;
    size_t sizeX, sizeY;
public:

const T& at(int x, int y) const { return data.at(y + x * sizeY); }

T& at(int x, int y) { return data.at(y + x * sizeY); }

// wrap other methods you need of std::vector here
};

Note that code above written on the fly and wasn't compiled. Still, I believe you've got the idea.

See also:

edited May 23 '17 at 10:31

Community

1
1

answered Sep 01 '14 at 19:20

Ivan Aksamentov - Drop

12,860
3
34
61

_"Risking to be downvoted for premature optimization, ..."_ Well I give you an initial charge ;). Your points are valid and relevant. – πάντα ῥεῖ Sep 01 '14 at 19:26
@Drop: You asserted that the performance in this case is terrible... So what is your suggestion for a efficient data structure for linear algebra and matrix manipulation?! Could you please explain that why your snippet might be better than 2D vectors, in view of the performance? – User Sep 01 '14 at 19:26
@matinking Well, that "terrible performance" is actually a bad and unprofessional way of describing my subjective understanding of how multidimensional arrays work. Surely you need to profile and think in a context of your application. Maybe vector of vectors will work just fine for you and you need no introduce additional complexity in your code. But, as a starting point I would advise to try to make a such a custom wrapper to `std:vector` and compare its performance to original approach. Also, me personally find it helpful in many applications, from tetris grid ;) to complex math apps. – Ivan Aksamentov - Drop Sep 01 '14 at 19:33
1

@matinking CPU cache is very important in our days so we must not throw away any possibility to abuse it (along with multithreading, vectorization and branch prediction). There are very good explanations given if you follow the "See also" links, so I would not like to repeat it. (And I am probably not able to explain it better) – Ivan Aksamentov - Drop Sep 01 '14 at 19:36
@Drop: +1 for the clearance and the resources :D, even though I still can't consider the other approach for my case!... – User Sep 01 '14 at 19:44

How to access to the index of a 2D vector by at() function?

2 Answers2

Linked