Any ideas how vectorise this?

Playing around with stock selection algorithm, hit this thousands of times any idea how to vectorise this. Any help appreciated.

function distances = calDistances(stocks, geneVector, dayN)
defaultNoData = 100;
distances(length(stocks)) = 100;

% Calculate Distances
for stockN = 1 : length(stocks)
    % Get Stock Vector
    stockVector =  stocks(stockN).getStockVector(dayN);
    % No Stock Data Check
    if not(isnan(stockVector))
        % Euclidian Distance
        distances(stockN) = norm(stockVector - geneVector);
    else
        % Default no data
        distances(stockN) = defaultNoData;
    end
end

[Jan]

What is the purpose of "not(isnan(stockVector))"? Do you want:

if all(not(isnan(stockVector)))

?

[Harry Smith]

Not all stocks have full data therefore to prevent an error in the euclidian distance I only use stocks with full data at that on that day, else default value

Answer 1

[Jan]

At first a simplified version:

function distances = calDistances(stocks, geneVector, dayN)
defaultNoData = 100;
distances     = repmat(defaultNoData, 1, length(stocks));

% Calculate Distances
for stockN = 1 : length(stocks)
  % Get Stock Vector
  stockVector = stocks(stockN).getStockVector(dayN);
  % No Stock Data Check
  if all(isfinite(stockVector))
      % Euclidian Distance
      distances(stockN) = norm(stockVector - geneVector);
  end
end

Use the profiler to check, if calling norm() uses a significant part of the time. If so, replace it by:

        distances(stockN) = sqrt(sum((stockVector - geneVector).^2));

While calculating the norm could be vectorized, this line cannot:

stockVector = stocks(stockN).getStockVector(dayN);

What is the class of stocks(stockN).getStockVector? Does getStockVector(dayN) really reply a vector?

Using a nested struct prevents a vectorization. You have to decide if the representation of the data should be nice to read or efficient to process.

[Harry Smith]

Thanks for the comment on euclidian distance, I'll have a look at that. Ive just checked the getStockVector, absolutely fine

The algorithm works fine just slow (24hours in a GA)