how velocize it? (it's possibile to vectorize it?)

Ntradess =[2    99     1     8    15    67    74    51    58    40
           3    80     7    14    16    73    55    57    64    41
           3    80    7     20    22    54    56    63    70    47
           3    81    7     21     3    60    62    69    71    28
           3    81    9     2      9    61    68    75    52    34
           4    82    9     83    90    42    49    26    33    65
           4    82    9     89    91    48    30    32    39    66
           5    85    10    95    97    29    31    38    45    72
           6    85    11    96    78    35    37    44    46    53
           7    85    12    77    84    36    43    50    27    58
           8    86    13    80    84    38    45    51    30    58
           8    86    13    82    84    38    45    51    31    58
           8    86    13    83    84    38    45    51    31    58
           9    87    13    84    84    38    45    51    31    58];

RP_bin=[0 0 1 0 0 1 0 0 1 0 0 1 0 0]';


period=2;
minTrades=2;
g=find(RP_bin>0);
Ntradess_=Ntradess(g,:);
[r,c]=size(Ntradess_);
   MinNtrad=zeros(r,c);
   for x=1:c %loop colonne
     for i=period+1:r 
         a=Ntradess_(i,x); %n:trade del idx attuale         
         if a>minTrades
           a=find(flip(Ntradess_(1:i-1,x))<=(a-minTrades),1);           
           if ~isempty(a) 
                MinNtrad(i,x)= i-a;  %MEMORIZZO la posizione della schiera per avere in mintrades richiesti

           end    
           
         end   
     end 
   end

[Dyuman Joshi]

What is your motivation behind wanting to vectorize the code above?

[pipin]

I think that encoded like this is slow

I use "find.". maybe it's not very efficient and it can be avoided (I ask)

[Walter Roberson]

You could save some coding effort if you were to use

a = find(Ntradess_(1:i-1,x))<=(a-minTrades), 1, 'last');
if ~isempty(a)
    MinNtrad(i,x) = a;
end

The change in logic might make it easier to figure out how to vectorize.

[pipin]

thank

Answer 1

[Bruno Luong]

I don't know the memory requirement would go up if I vectorize the outer loop in a non-toy case. So I leave it for now.

Ntradess =[2    99     1     8    15    67    74    51    58    40
           3    80     7    14    16    73    55    57    64    41
           3    80    7     20    22    54    56    63    70    47
           3    81    7     21     3    60    62    69    71    28
           3    81    9     2      9    61    68    75    52    34
           4    82    9     83    90    42    49    26    33    65
           4    82    9     89    91    48    30    32    39    66
           5    85    10    95    97    29    31    38    45    72
           6    85    11    96    78    35    37    44    46    53
           7    85    12    77    84    36    43    50    27    58
           8    86    13    80    84    38    45    51    30    58
           8    86    13    82    84    38    45    51    31    58
           8    86    13    83    84    38    45    51    31    58
           9    87    13    84    84    38    45    51    31    58];

RP_bin=[0 0 1 0 0 1 0 0 1 0 0 1 0 0]';

period=2;
minTrades=2;
g=find(RP_bin>0);
Ntradess_=Ntradess(g,:);

[r,c]=size(Ntradess_);
MinNtrad=zeros(r,c);
for x=1:c %loop colonne
    for i=period+1:r
        a=Ntradess_(i,x); %n:trade del idx attuale
        if a>minTrades
            a=find(flip(Ntradess_(1:i-1,x))<=(a-minTrades),1);
            if ~isempty(a)
                MinNtrad(i,x)= i-a;  %MEMORIZZO la posizione della schiera per avere in mintrades richiesti
            end

        end
    end
end
MinNtrad

MinNtrad = 4×10

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 2 2 1 0 0 2 2 1 3 2 3 1 3 3 3 3 0 3

[r,c]=size(Ntradess_);
MinNtrad = zeros(r,c);
Mask = flip(triu(true(r),1),1);
for x = 1:c
    nx = Ntradess_(:,x);
    [v,j] = max(Mask .* (nx'-flip(nx,1) >= minTrades), [], 1);
    j = r+1-j;
    j(v(:)==0 | nx(:)<=minTrades) = 0;
    MinNtrad(:,x) = j;
end
MinNtrad(1:period,:) = 0;
MinNtrad

MinNtrad = 4×10

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 2 2 1 0 0 2 2 1 3 2 3 1 3 3 3 3 0 3

[Bruno Luong]

Full vectorize

warning: potential runout of memory

[r,c] = size(Ntradess_);
Mask = (1:r)' > (r:-1:1); % flip(triu(true(r),1),1);
D = reshape(Ntradess_, [1 r c]) - reshape(flip(Ntradess_, 1), [r 1 c]);
B = Mask .* (D >= minTrades);
[V,J] = max(B, [], 1);
MinNtrad = r+1-reshape(J, [r c]);
V = reshape(V, [r c]);
MinNtrad(V==0 | Ntradess_<=minTrades) = 0;
MinNtrad(1:period,:) = 0;

[pipin]

congrats for the vectorization....but i don't use it it requires a lot of free memory