Seek: more efficient implementation

Hi,

I have a function H2_GFL (code below) which correctly performs the task I want it to do.

- (create a 3-dimensional cell; I will fill the the first dim with vectors later, at runtime; 2nd dim will contain random numbers bewtween 0 and 1.).

Is there any other code implementation by which this functionality can be done more efficiently? In my application I shall be calling (invoking) this function millions of times with N ranging from 20 to 300 and max_periods ranging from 1000 to 10,000. I use Matlab 7.01.24104 (R14, SP1) and R2015b (64-bit).

Regards

Sasanka

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

function H_L1 = H2_GFL(N, max_periods)

num_rows = max_periods * N;

H_L1 = cell(num_rows, 2, N);

rand_supply = rand(num_rows, N);

for k = 1:1:N

for j = 1:1:num_rows

H_L1(j,2,k) = {rand_supply(j,k)};

H_L1(j,1,k) = {-98}; %% Dummy value, indicates end of data row

end;

end;

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

Results with N = 3, max_periods = 5;

H_L1(:,:,1) =

[-98] [0.2715]

[-98] [0.4673]

[-98] [0.1713]

[-98] [0.7865]

[-98] [0.8774]

[-98] [0.0234]

[-98] [0.6326]

[-98] [0.5651]

[-98] [0.5763]

[-98] [0.8755]

[-98] [0.4644]

[-98] [0.7747]

[-98] [0.5460]

[-98] [0.7966]

[-98] [0.1721]

H_L1(:,:,2) =

[-98] [0.1791]

[-98] [0.5115]

[-98] [0.6513]

[-98] [0.4854]

[-98] [0.2452]

[-98] [0.5454]

[-98] [0.3246]

[-98] [0.9038]

[-98] [0.1873]

[-98] [0.5310]

[-98] [0.2923]

[-98] [0.9244]

[-98] [0.8981]

[-98] [0.1892]

[-98] [0.7519]

H_L1(:,:,3) =

[-98] [0.8209]

[-98] [0.9932]

[-98] [0.5719]

[-98] [0.9268]

[-98] [0.7205]

[-98] [0.7902]

[-98] [0.3668]

[-98] [0.8995]

[-98] [0.2396]

[-98] [0.7576]

[-98] [0.7577]

[-98] [0.4745]

[-98] [0.9994]

[-98] [0.6355]

[-98] [0.8145]

[Stephen23]

Using a cell array slows things down... why not just return a numeric array?

[Sasanka Sekhar Chanda]

I wish to store vectors in the first colum. Regards Sasanka

[Sasanka Sekhar Chanda]

Dear Stephen,

Your tip had me thinking:

Actually, I could use a numeric array (instead of cell), if I give up on requiring the first column to have vectors.

The vectors I have in mind are row vectors with values 0 and 1, having (possibly) at most 300 to 1000 members.

I could potentially convert the vector into 1 number (of upto 1000 digits) by concatenating the members , as given by the code in below link.

https://in.mathworks.com/matlabcentral/answers/307683-concatenate-all-value-in-a-matrix-into-one-number

col_1_member = sum(10.^(numel(config_v)-1:-1:0) .* config_v);

Question in my mind is: Am I protected from rounding errors? Will matlab correctly distinguish between a very large number and another number which is the previous number + 1? Also, If I use above code to concatenate vector members many times would there by a large(r) performance penalty?

Regards

Sasanka

[Stephen23]

"Am I protected from rounding errors?"

Not with 1000 digits, that is well beyond the precision of any native numeric type. And converting to-from the digit representations will not be particularly efficient over your millions of iterations.

Would it be possible to return two outputs?:

1. a cell array of numeric vectors
2. a numeric array of random numbers

[Sasanka Sekhar Chanda]

Dear Stephen,

I think what you say should be do-able in my situation.

(a). In the separate cell array of numeric vectors I shall keep my vectors. Cell (1,1) will have the first vector; Cell (1,2) will have the second vector; etc.

(b). I will have a separate (numeric) matrix with 2 columns (and N*max_periods rows): the second column shall have random numbers. The first col. shall have -98 in all rows at the beginning.

(c). At runtime, I will refer to the cell index number for the vector of my interest (in the separate cell array of numeric vectors - the numbers in bold 1, 2 above), and insert this row number in the first col. of the numeric array (replacing the -98 dummy value). This shall enable me to refer back, should I start from the numeric matrix and wish to know the vector a row (of tha numeric matrix ) corresponded to.

Thank you v. v. much!

Regards

Sasanka

Answer 1

[KSSV]

function H_L1 = H2_GFL(N, max_periods)
num_rows = max_periods * N;
H_L1 = zeros(num_rows, 2, N); 
rand_supply = rand(num_rows, N);
for k = 1:1:N   
        H_L1(:,2,k) = rand_supply(:,k);
        H_L1(:,1,k) =  -98; %% Dummy value, indicates end of data row     
end

No loop required.

function H_L1 = H2_GFL(N, max_periods)
num_rows = max_periods * N;
H_L1 = zeros(num_rows, 2, N); 
rand_supply = rand(num_rows, N);
H_L1(:,2,:) = rand_supply ; 
H_L1(:,1,:) = -98 ; 
end

[Sasanka Sekhar Chanda]

Dear KSSV,

Thank you for the suggestion.

In the first column I shall be storing very large numbers or vectors. So, only using a numeric array is ruled out.

I am trying out the suggestion given by Stephen Cobeldick above. That may solve the problem.

Have a nice day!

Regards

Sasanka