Please help. I'm new to matlab scripting and need a bit of help. I have a series of numbers:
A=[1 1 1 2 2 2 3 3 3 4 4 4 5]
which I want to fill randomly into an 8x12 matrix without having the same number in the same row. At the end I want all the "empty" cells of the 8x12 matrix being filled with 0's or nan.
an example could be:
result=
3 1 5 2 4 5 0 0 0 0 0 0
4 1 3 2 0 0 0 0 0 0 0 0
1 3 4 2 0 0 0 0 0 0 0 0
make sure A is sorted. A = sort(A)more info
make an empty matrix.
For each number in A: more info
find out how many repetitions of the number there are -> for loop in A, start is the first occurance of the number, end is the last, n = last-first+1
find all rows that have space for an extra number, just do a double for loop and keep track of elements that are zero
randomly pick n rows -> more info. To do this, make an array R of all available row indixes. Then take a random sample between 1..size(R,2) with the provided function and get all the values, you now have your row indixes.
randomly pick one of the empty spots in each of the selected rows and assign the number
Related
In APL, matrices and vectors are used to hold data. I was wondering if there was a way to search within a matrix for a given value, and have that values index returned. For example, say I have the following 2-dimensional matrices:
VALUES ← 1 2 3 4 5 6 7 8 9 10 11... all the way up to 36
KINDS ← 0 0 0 2 0 0 0 3 0 ... filled with 0's the rest of the way to 36 length.
If I laminated these two matrices with
kinds,[.5] values
so that they are laminated one on top of the other
1 2 3 4 5 6 7 8 9 10...
0 0 0 2 0 0 0 3 0 ....
is there a functionally easy way to search for the index of the 2 value in the "second row" of the newly laminated matrix? eg. the column containing
4
2
and return that matrix index?
The value 2 also appears in row 1 of your newly laminated matrix (nlm), and as you stated, you really do not want to search the whole matrix, but only the second row. So, since you're only searching within a given row, getting the column index in that row gives you the complete answer:
row←2
⎕←col←nlm[row;]⍳2
4
nlm[;col] ⍝ values in matched column
4 2
Try it online!
I have the following problem:
Given a matrix A
A = [ 1 2 2 3 3 ;
2 2 2 7 9 ]
where the sequence of unique numbers within the matrix is not continuous. In this example
unique(A) = [ 1 2 3 7 9 ]. % [ 4 5 6 8 ] are missing
I want to compute the same matrix, but using instead a continuous sequence, such that
unique(A_new) = [ 1 2 3 4 5 ];
I came up with the following solution
T = [ unique(A), [ 1:numel(unique(A)) ]' ];
A_new = zeros(size(A));
for i = 1:size(T,1)
A_new( A == T(i,1) ) = T(i,2);
end
This is incredibly slow: the size of the matrix A I have to work with is 200x400x300 and the the number of unique elements within this matrix is 33406.
Any idea on how to speed up the procedure?
If I understand correctly, in your example you want:
A_new = [ 1 2 2 3 3 ;
2 2 2 4 5 ]
So just compute a lookup table (lookup) such that you can then do:
A_new = lookup(A);
So in your case, lookup would be:
[ 1 2 3 0 0 0 4 0 5 ]
I'll leave the process for generating that as an exercise for the reader.
Approach 1 (not recommended)
This should be pretty fast, but it uses more memory:
[~, A_new] = max(bsxfun(#eq, A(:).', unique(A(:))));
A_new = reshape(A_new, size(A));
How does this work?
First A is linearized into a vector (A(:)). Also, a vector containing the unique values of A is computed (unique(A(:))). From those two vectors a matrix is generated (with bsxfun) in which each entry of A is compared to each of the unique values. That way, for each entry of A we know if it equals the first unique value, or the second unique value, etc. For the A given in your question, that matrix is
1 0 0 0 0 0 0 0 0 0
0 1 1 1 1 1 0 0 0 0
0 0 0 0 0 0 1 0 1 0
0 0 0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 0 0 1
So for example, the value 1 in entry (2,3) indicates that the third value of A(:) equals the second unique value of A (namely 2). The 1 in the lower-right entry (5,10) indicates that the tenth value of A(:) is the fifth unique value of A (which is 9).
Now the second output of max is used to extract the row position of the 1 value in each columnn (i.e. to obtain the numbers indicating "second", "fifth" etc in the above example)). These are the desired results. It only remains to reshape them into the shape of A.
Approach 2 (recommended)
The third output of unique does what you want:
[~, ~, labels] = unique(A);
A_new = reshape(labels, size(A));
I was recently reading about Lehmer codes and how they can be used to map an index to a permutation corresponding to that index and realized they can be quite useful to quickly generate a permutation. Does anyone know how can this be done using an algorithm, and also what are the limits of such a method, I suppose we can't go above index = 1.7977e+308, but still seems quite an interesting method.
So basically lets say we have
perm
1 0 0 0
2 0 0 1
3 0 0 2
4 0 1 0
5 0 1 1
6 0 1 2
...
We should be able to deduce that the index of [ 0 1 0 ] is 4,
or that the index 6 corresponds to [ 0 1 2 ]
Thanks for any help.
The vector for each index is the base 3 representation of the index (minus one)
the functions dec2base and base2dec can be used for this with a little fiddling to get the sting outputs to the required format
index to vector
index=4; % input index
n=3; % length of vector
vec=str2num([dec2base(index-1,3,n)].').'
vec=
0 1 0
vector to index
vec=[0,1,2]; % input vector
vecstr=strcat(['' vec(:)+'0'].');
index=base2dec(vecstr,3)+1
index =
6
I've been searching for an algorithm for the solution of all possible matrices of dimension 'n' that can be obtained with two arrays, one of the sum of the rows, and another, of the sum of the columns of a matrix. For example, if I have the following matrix of dimension 7:
matriz= [ 1 0 0 1 1 1 0
1 0 1 0 1 0 0
0 0 1 0 1 0 0
1 0 0 1 1 0 1
0 1 1 0 1 0 1
1 1 1 0 0 0 1
0 0 1 0 1 0 1 ]
The sum of the columns are:
col= [4 2 5 2 6 1 4]
The sum of the rows are:
row = [4 3 2 4 4 4 3]
Now, I want to obtain all possible matrices of "ones and zeros" where the sum of the columns and the rows fulfil the condition of "col" and "row" respectively.
I would appreciate ideas that can help solve this problem.
One obvious way is to brute-force a solution: for the first row, generate all the possibilities that have the right sum, then for each of these, generate all the possibilities for the 2nd row, and so on. Once you have generated all the rows, you check if the sum of the columns is right. But this will take a lot of time. My math might be rusty at this time of the day, but I believe the number of distinct possibilities for a row of length n of which k bits are 1 is given by the binomial coefficient or nchoosek(n,k) in Matlab. To determine the total number of possibilities, you have to multiply this number for every row:
>> n = 7;
>> row= [4 3 2 4 4 4 3];
>> prod(arrayfun(#(k) nchoosek(n, k), row))
ans =
3.8604e+10
This is a lot of possibilities to check! Doing the same for the columns gives
>> col= [4 2 5 2 6 1 4];
>> prod(arrayfun(#(k) nchoosek(n, k), col))
ans =
555891525
Still a large number, but 'only' a factor 70 smaller.
It might be possible to improve this brute-force method a little bit by seeing if the later rows are already constrained by the previous rows. If in your example, for a particular combination of the first two rows, both rows have a 1 in the second column, the rest of this column should all be 0, since the sum must be 2. This reduces the number of possibilities for the remaining rows a bit. Implementing such checks might complicate things a bit, but they might make the difference between a calculation that takes 2 days or one that takes just 1 hour.
An optimized version of this might alternatively generate rows and columns, and start with those for which the number of possibilities is the lowest. I don't know if there is a more elegant solution than this brute-force method, I would be interested to hear one.
Is it possible to decompose a matrix A having n rows and n columns to sum of m [n x n] permutation matrices. where m is the number of 1's in each row and each column in matrix A?
UPDATE:
yes, this is possible. I came across such an exmaple which is shown below - but How can we generalize the answer?
What you want is called a 1-factorization. One algorithm is repeatedly to find a perfect matching and remove it; probably there are others.
For the first permutation matrix, take the first 1 in the first row. For the second row, take the first 1 that is in a column you don't already have. For the third row, take the first 1 that is in a column you don't already have. And so on. Do this for all rows.
You now have one permutation matrix.
Next subtract your first permutation matrix from the original. This new matrix now has m-1 ones in each row and column. So repeat the process m-1 more times, and you'll have your m permutation matrices.
You can skip the last step, because a matrix with one 1 in each row and column already is a permutation matrix. There's no need to do any calculations.
This is a greedy algorithm that doesn't always work. We can make it work by changing the selection rule slightly. See below:
For your example:
1 0 1 1
A = 1 1 0 1
1 1 1 0
0 1 1 1
In the first step, we pick (1,1) for the first row, (2,2) for the second row, (3,3) for the thrid row and (4,4) for the 4th row. We then have:
1 0 0 0 0 0 1 1
A = 0 1 0 0 + 1 0 0 1
0 0 1 0 1 1 0 0
0 0 0 1 0 1 1 0
The first matrix is a permutation matrix. The second matrix has exactly two 1's in each row and column. So we pick, in order: (1,3), (2,1), (3,2) and... we're in trouble: the rows that contain a 1 in column 4 have already been used.
So how do we fix this? Well, we can keep track of the number of 1's remaining in each column. Instead of picking the first column that is unused, we pick the column with the lowest number of 1's remaining. For the second matrix above:
0 0 1 1 0 0 X 0 0 0 X 0 0 0 X 0
B = 1 0 0 1 --> 1 0 0 1 --> 0 0 0 X --> 0 0 0 X
1 1 0 0 1 1 0 0 1 1 0 0 X 0 0 0
0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0
------- ------- ------- -------
2 2 2 2 2 2 X 1 1 2 X X X 1 X X
So we would pick column 4 in the second step, column 1 in the 3rd step, and column 2 in the 4th step.
There can always be only one column with one remaining 1. The other 1's must have been taken away in m-1 previous rows. If you had two such columns, one of them would have had to have been picked as the minimum column before.
This can be done easily using a recursive (backtracking OR depth-first traversal) algorithm. Here is the pseudo-code for its solution:
void printPermutationMatrices(const int OrigMat[][], int permutMat[], int curRow, const int n){
//curPermutMatrix is 1-D array where value of ith element contains the value of column where 1 is placed in ith row
if(curRow == n){//Base case
//do stuff with permutMat[]
printPermutMat(permutMat);
return;
}
for(int col=0; col<n; col++){//try to place 1 in cur_row in each col if possible and go further to next row in recursion
if(origM[cur_row][col] == 1){
permutMat[cur_row] = col;//choose this col for cur_row
if there is no conflict to place a 1 in [cur_row, col] in permutMat[]
perform(origM, curPermutMat, curRow+1, n);
}
}
}
Here is how to call from your main function:
int[] permutMat = new int[n];
printPermutationMatrices(originalMatrix, permutMat, 0, n);