I would like to do simple stuff - extract every second row in a matrix of shape [4,5] to get two output matrices of shapes [2,5]. My matrix is:
Eigen::MatrixXf m(4,5);
m << 1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20;
when I print I get the expected:
1 2 3 4 5
6 7 8 9 10
11 12 13 14 15
16 17 18 19 20
Now how to obtain the expected outputs: ?
// expected mOut1:
1 2 3 4 5
11 12 13 14 15
// and expected mOut2:
6 7 8 9 10
16 17 18 19 20
So far I tried with Eigen::Map and strides, but I have no idea how do those strides work.
I made a function:
Eigen::Map<Eigen::MatrixXf, 0, Eigen::Stride<Eigen::Dynamic, Eigen::Dynamic>> myMatrixSlice(const float* data, unsigned dim1, unsigned dim2, int stride1, int stride2) {
using Stride = Eigen::Stride<Eigen::Dynamic, Eigen::Dynamic>;
return Eigen::Map<Eigen::MatrixXf, 0, Stride>{
const_cast<float*>(data), dim1, dim2, Stride{stride1, stride2}};
}
and I want to call it like this:
auto mOut1 = myMatrixSlice(m.data(), 2, 5, ?, ?);
auto mOut2 = myMatrixSlice(m.data(), 2, 5, ?, ?); // remember to start with row=1
but I guess I need some help on this.
Related
so i need to swap row 2 and row 3 of my array. There is a catch our professor had us use a one dimensional array and use a pointer instead of a 2d array. We cannot reference the array only the pointer. I have no idea how to do this.
int numbers[25] = { 1,3,5,7,9 , -2,-4,-6, -8, -10 , 3,3,3,3,3 , 55, 77, 99, 22, 33, -15, -250, -350, -450, -550 };
this array is supposed to look like this:
1 3 5 7 9
-2 -4 -6 -8 -10 // i need to swap this row
3 3 3 3 3 // for this row
55 77 99 22 33
-15 -250 -350 -450 -550
This is how i need to print it
1 3 5 7 9
3 3 3 3 3
-2 -4 -6 -8 -10
55 77 99 22 33
-15 -250 -350 -450 -550
Note: this is not my whole HW assignment is just where i am stuck.
Why not try something like:
constexpr std::size_t rowLength = 5u;
const auto beginRow2 = std::begin(numbers) + (rowLength * 2);
const auto endRow2 = std::begin(numbers) + (rowLength * 3);
const auto beginRow3 = std::begin(numbers) + (rowLength * 3);
std::swap_ranges(beginRow2, endRow2, beginRow3);
This is idiomatic C++ and can easily be adapted to provide a generic function that accepts a 1-dimensional container, a row-length and the two rows to swap.
Just define a temporary array:
int tmp_row[5];
Save the third row:
int bytes = sizeof(tmp_row);
memcpy(tmp_row, &numbers[10], bytes);
Then fill in the second and third rows appropriately:
memcpy(&numbers[10], &numbers[5], bytes);
memcpy(&numbers[5], tmp_row, bytes);
I'm trying to use the MKL routine mkl_dcsradd to add an upper-triangular matrix to its transpose. In this case, the upper triangular matrix stores part of the adjacency matrix of a graph, and I need the full version for implementing another algorithm.
In this simplified example, I start with a list of (11) edges, and build an upper-triangular CSR matrix from it. I have checked that this much works. However, when I try to add it to its transpose, dcsradd stops on the final row, saying it's run out of space. However, this shouldn't be the case. An upper triangular matrix (no zeros along the diagonal) with n non-zero entries, when added to its transpose, should result in a matrix with 2n (22) non-zeros.
When I supply dcsradd with a maximum non-zeros of 22, it fails, but when I supply it with 23 (an excessive value), it works correctly. Why is this?
I've simplified my code down to a minimal example demonstrating the error:
#include <cstdint>
#include <cstdio>
#include <cstdlib>
#include <mkl.h>
int main()
{
int nnz = 11;
int numVertices = 10;
int32_t u[] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 0, 1 };
int32_t v[] = { 1, 2, 3, 4, 5, 6, 7, 8, 9, 5, 8 };
double w[] = { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 };
int fullNnz = nnz * 2;
int dim = numVertices;
double triData[nnz];
int triCols[nnz];
int triRows[dim];
// COO to upper-triangular CSR
int info = -1;
int job [] = { 2, 1, 0, 0, nnz, 0 };
mkl_dcsrcoo(job, &dim,
triData, triCols, triRows,
&nnz, w, u, v,
&info);
printf("info = %d\n", info);
// Allocate final memory
double data[fullNnz];
int cols[fullNnz];
int rows[dim];
// B = A + A^T (to make a full adjacency matrix)
int request = 0, sort = 0;
double beta = 1.0;
int WRONG_NNZ = fullNnz + 1; // What is happening here?
mkl_dcsradd("t", &request, &sort, &dim, &dim,
triData, triCols, triRows,
&beta, triData, triCols, triRows,
data, cols, rows,
&WRONG_NNZ, &info);
printf("info = %d\n", info);
// Convert back to 0-based indexing (via Cilk)
cols[:]--;
rows[:]--;
printf("data:");
for (double d : data) printf("%.0f ", d);
printf("\ncols:");
for (int c : cols) printf("%d ", c);
printf("\nrows:");
for (int r : rows) printf("%d ", r);
printf("\n");
return 0;
}
I compile with:
icc -O3 -std=c++11 -xHost main.cpp -o main -openmp -L/opt/intel/composerxe/mkl/lib -lmkl_intel_lp64 -lmkl_core -lmkl_intel_thread -lpthread -lm
When I give 22, the output is:
info = 0
info = 10
data:1 10 1 2 11 2 3 3 4 4 5 10 5 6 6 7 7 8 11 8 9 0
cols:1 5 0 2 8 1 3 2 4 3 5 0 4 6 5 7 6 8 1 7 9 -1
rows:0 2 5 7 9 11 14 16 18 21
But, when I give 23, the output is:
info = 0
info = 0
data:1 10 1 2 11 2 3 3 4 4 5 10 5 6 6 7 7 8 11 8 9 9
cols:1 5 0 2 8 1 3 2 4 3 5 0 4 6 5 7 6 8 1 7 9 8
rows:0 2 5 7 9 11 14 16 18 21
Assume that I have a vector:
x = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14]
What I need to do is split this vector into block sizes of blocksize with an overlap
blocksize = 4
overlap = 2
The result, would be a 2D vector with size 4 containing 6 values.
x[0] = [1, 3, 5, 7, 9, 11]
x[1] = [ 2 4 6 8 10 12]
....
I have tried to implement this with the following functions:
std::vector<std::vector<double> > stride_windows(std::vector<double> &data, std::size_t
NFFT, std::size_t overlap)
{
std::vector<std::vector<double> > blocks(NFFT);
for(unsigned i=0; (i < data.size()); i++)
{
blocks[i].resize(NFFT+overlap);
for(unsigned j=0; (j < blocks[i].size()); j++)
{
std::cout << data[i*overlap+j] << std::endl;
}
}
}
This is wrong, and, segments.
std::vector<std::vector<double> > frame(std::vector<double> &signal, int N, int M)
{
unsigned int n = signal.size();
unsigned int num_blocks = n / N;
unsigned int maxblockstart = n - N;
unsigned int lastblockstart = maxblockstart - (maxblockstart % M);
unsigned int numbblocks = (lastblockstart)/M + 1;
std::vector<std::vector<double> > blocked(numbblocks);
for(unsigned i=0; (i < numbblocks); i++)
{
blocked[i].resize(N);
for(int j=0; (j < N); j++)
{
blocked[i][j] = signal[i*M+j];
}
}
return blocked;
}
I wrote this function, thinking that it did the above, however, it will just store:
X[0] = 1, 2, 3, 4
x[1] = 3, 4, 5, 6
.....
Could anyone please explain how I would go about modifying the above function to allow for skips by overlap to take place?
This function is similar to this: Rolling window
EDIT:
I have the following vector:
1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14
I want to split this vector, into sub-blocks (thus creating a 2D vector), with an overlap of the parameter overlap so in this case, the parameters would be: size=4 overlap=2, this would then create the following 2D vector:
`block0 = [ 1 3 5 7 9 11]
block1 = [ 2 4 6 8 10 12]
block2 = [ 3 5 7 9 11 13]
block3 = [ 4 6 8 10 12 14]`
So essentially, 4 blocks have been created, each block contains a value where the element is skipped by the overlap
EDIT 2:
This is where I need to get to:
The value of overlap will overlap the results of x in terms of placements inside the vector:
block1 = [1, 3, 5, 7, 9, 11]
Notice from the actual vector block:
1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14
Value: 1 -> This is pushed into block "1"
Value 2 -> This is not pushed into block "1" (overlap is skip 2 places in the vector)
Value 3 -> This is pushed into block "1"
value 4 -> This is not pushed into block "1" (overlap is skip to places in the vector)
value 5 -> This is pushed into block "1"
value 6 -> "This is not pushed into block "1" (overlap is skip 2 places in the vector)
value 7 -> "This value is pushed into block "1"
value 8 -> "This is not pushed into block "1" (overlap is skip 2 places in the vector)"
value 9 -> "This value is pushed into block "1"
value 10 -> This value is not pushed into block "1" (overlap is skip 2 places in the
vector)
value 11 -> This value is pushed into block "1"
BLOCK 2
Overlap = 2;
value 2 - > Pushed back into block "2"
value 4 -> Pushed back into block "2"
value 6, 8, 10 etc..
So each time, the place in the vector is skipped by the "overlap" in this case, it is the value of 2..
This is what the expected output would be:
[[ 1 3 5 7 9 11]
[ 2 4 6 8 10 12]
[ 3 5 7 9 11 13]
[ 4 6 8 10 12 14]]
If I understand you correctly, you're pretty close. You need something like the following. I used int because frankly its easier to type than double =P
#include <iostream>
#include <algorithm>
#include <vector>
#include <limits>
#include <iterator>
std::vector<std::vector<int>>
split(const std::vector<int>& data, size_t blocksize, size_t overlap)
{
// compute maximum block size
std::vector<std::vector<int>> res;
size_t minlen = (data.size() - blocksize)/overlap + 1;
auto start = data.begin();
for (size_t i=0; i<blocksize; ++i)
{
res.emplace_back(std::vector<int>());
std::vector<int>& block = res.back();
auto it = start++;
for (size_t j=0; j<minlen; ++j)
{
block.push_back(*it);
std::advance(it,overlap);
}
}
return res;
}
int main()
{
std::vector<int> data { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 };
for (size_t i=2; i<6; ++i)
{
for (size_t j=2; j<6; ++j)
{
std::vector<std::vector<int>> blocks = split(data, i, j);
std::cout << "Blocksize = " << i << ", Overlap = " << j << std::endl;
for (auto const& obj : blocks)
{
std::copy(obj.begin(), obj.end(), std::ostream_iterator<int>(std::cout, " "));
std::cout << std::endl;
}
std::cout << std::endl;
}
}
return 0;
}
Output
Blocksize = 2, Overlap = 2
1 3 5 7 9 11 13
2 4 6 8 10 12 14
Blocksize = 2, Overlap = 3
1 4 7 10 13
2 5 8 11 14
Blocksize = 2, Overlap = 4
1 5 9 13
2 6 10 14
Blocksize = 2, Overlap = 5
1 6 11
2 7 12
Blocksize = 3, Overlap = 2
1 3 5 7 9 11
2 4 6 8 10 12
3 5 7 9 11 13
Blocksize = 3, Overlap = 3
1 4 7 10
2 5 8 11
3 6 9 12
Blocksize = 3, Overlap = 4
1 5 9
2 6 10
3 7 11
Blocksize = 3, Overlap = 5
1 6 11
2 7 12
3 8 13
Blocksize = 4, Overlap = 2
1 3 5 7 9 11
2 4 6 8 10 12
3 5 7 9 11 13
4 6 8 10 12 14
Blocksize = 4, Overlap = 3
1 4 7 10
2 5 8 11
3 6 9 12
4 7 10 13
Blocksize = 4, Overlap = 4
1 5 9
2 6 10
3 7 11
4 8 12
Blocksize = 4, Overlap = 5
1 6 11
2 7 12
3 8 13
4 9 14
Blocksize = 5, Overlap = 2
1 3 5 7 9
2 4 6 8 10
3 5 7 9 11
4 6 8 10 12
5 7 9 11 13
Blocksize = 5, Overlap = 3
1 4 7 10
2 5 8 11
3 6 9 12
4 7 10 13
5 8 11 14
Blocksize = 5, Overlap = 4
1 5 9
2 6 10
3 7 11
4 8 12
5 9 13
Blocksize = 5, Overlap = 5
1 6
2 7
3 8
4 9
5 10
I have a matrix in SAS/IML:
x = {7 6 3 3 8,
2 3 5 2 5,
2 6 4 3 8,
7 4 8 1 3,
8 8 6 8 7,
3 2 6 1 5 };
I want to create a new matrix that contains the highest k values of each column in x. For example, if k=3, I want the result matrix to contain:
8 8 8 8 8
7 6 6 3 8
7 6 6 3 7
because, for instance, the largest 3 numbers in the first column of x are 8, 7, and 7.
I've unsuccessfully tried to figure out how to do this using the rank function.
Your code looks fine. Here's a minor revision:
do c=1 to ncol(x);
r = rank(x[,c]);
y = x[loc(r>=nrow(x)-k+1), c];
call sort(y);
tops[,c] = y;
end;
As to avoiding the loop to make it faster, it's not necessary. Even with 10,000 columns, this code runs in a fraction of a second. Try running the following timing code:
x = j(500, 10000);
call randgen(x,"normal");
k = 3;
t0=time();
tops = j(k,ncol(x),0);
do c=1 to ncol(x);
r = rank(x[,c]);
y = x[loc(r>=nrow(x)-k+1), c];
call sort(y);
tops[,c] = y;
end;
t=time()-t0;
print t;
Here's a partial answer I've come up with:
k = 3;
tops = j(k,ncol(x),0);
do c=1 to ncol(x);
r = rank(x[,c]);
h=loc(r>=nrow(x)-k+1);
tops[,c] = x[,c][h];
end;
This approach uses a loop, which I'd like to avoid, so please post improvements if possible!
How to pick the best uniformed 1d array from the 2d arrays ?
I have two 2d array of : 11 x 10
Example :
4 8 12 12 12 14 16 18 4 1 0
5 7 11 12 13 11 15 18 3 2 1
8 3 12 14 18 19 20 21 8 5 4 ,
8 2 11 12 17 17 19 20 7 4 3 ,
4 7 11 11 11 15 17 19 5 1 1 ,
3 8 11 13 11 15 14 17 4 1 0 ,
4 7 12 13 13 14 16 19 3 1 1 ,
5 9 11 12 13 15 17 19 5 0 1 ,
9 7 25 22 24 18 23 17 3 3 3 ,
4 8 13 13 13 15 17 17 5 2 0 ,
here we have 2d arrays of size 11x10 - Need to analysis and have to find out the common 1d array which has common like.
find the best closing number and its difference- and keep doing for all the corresponding columns in an array .
below answer should be like - finding the corresponding very column and comparing with the next row column - if it has some difference below ( 5 ) take the both column of two rows are same and process for next column of the same row..process untill finding the 1 row where it has at least nearby matches of 5
4 8 11 12 13 13 15 18 4 1 0
why don't you do something like this
int[] count(int[][] array)
int result[11];
for(int x = o; x<= 11;x++)
{
int temp[10];
for(int y = o; y<= 10;y++)
{
temp[y] = array[y][x];
}
result[x] = getHighest(temp);
}
return result;
}
int getHighest(int[] array)
{
int size = array.length;
int[size][1] temp;
for(int x; x<= size;x++)
{
int z = array[x];
temp[z][0]++;
}
int highest = -1;
for(int z; z<= size;z++)
{
int z = array[x];
int h = temp[z][0];
if(h > highest)
{
highest = h;
}
}
return highest;
}
Something like this, but my C++ has gotten a bit of rusty so sorry if there are any mistakes.