turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

Topic Options

- RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

08-31-2011 05:01 PM

hey,

I have a piece of matlab code to generate normally distributed cluster data. Since I don't have matlab installed in my PC, I am thinking to translate it to SAS IML. I just know a little with matlab. could anyone help with this? The matlab code is as below. Thanks.

%

rand('state',91225);

randn('state',19481);

lowBound = -50;

highBound =50;

nCenters = 20;

nCols = 32;

nRows = 20000;

nTestRows = 0.01 * nRows;

nBufferPoints = 100000;

nExpandFactor = 10; % How much to stretch the covariance matrix

sTrainFile = 'outtrain.txt';

sTestFile = 'outtest.txt';

% Generate the centers according to a uniform distibution.

mCenters = round(lowBound + rand(nCenters,nCols)*(highBound-lowBound));

% Generate the variances and covariances randomly to create a matrix for

% each center

mCovariance = zeros(nCols,nCols);

cCovariance = cell(nCenters,1);

for i = 1:nCenters,

mRootCovariance = nExpandFactor * ...

rand(nCols,nCols)*(highBound-lowBound) / 50;

cCovariance{i} = mRootCovariance' * mRootCovariance;

end;

% Determine what proportion of points will come from each center, then

% create a cdf to use in deciding which to generate.

vPointFraction = rand(nCenters,1);

vPointFraction = vPointFraction / sum(vPointFraction);

vPointCdf = zeros(1,nCenters);

for i = 1:nCenters,

vPointCdf(i) = sum(vPointFraction(1:i));

end;

% Create a random separating plane.

w = -2 + rand(nCols,1)*4;

gamma = lowBound / 10 + rand * (highBound-lowBound)/10;

% Now choose which classes to which each center belongs

vCenterClasses = sign(mCenters * w - gamma * ones(nCenters,1));

vZeroSpots = find(vCenterClasses==0);

vCenterClasses(vZeroSpots) = ones(length(vZeroSpots),1);

% Prepare output file

flatfile([],sTrainFile,0);

flatfile([],sTestFile,0);

% Now go through and begin generating random points.

% Do it twice: once for testing, once for training.

for nDataset = 1:2,

if (nDataset==1)

nRowsLeft = nRows;

sOutputFile = sTrainFile;

nTotRows = nRows;

else

nRowsLeft = nTestRows;

sOutputFile = sTestFile;

nTotRows = nTestRows;

end;

nMisclass = 0;

nTrainingClass1 = 0;

nTrainingClassm1 = 0;

while (nRowsLeft > 0)

disp(sprintf('Rows left = %d',nRowsLeft));

nRowsNow = min(nBufferPoints,nRowsLeft);

nRowsLeft = nRowsLeft - nRowsNow;

mNewPoints = zeros(nRowsNow,nCols);

vPointCenters = zeros(nRowsNow,1);

% Determine which center each point should belong to

vRandomNumbers = rand(nRowsNow,1);

for i = nCenters:-1:1,

vCenterMatch = (vRandomNumbers <= vPointCdf(i));

vPointCenters([vCenterMatch]) = i;

end;

% Create a vector of training classes for each point

vTrainingClasses = zeros(nRowsNow,1);

% Within each class, generate an appropriate number of random points.

for i = 1:nCenters,

vIndices = (vPointCenters==i);

nPoints = sum(vIndices);

vTrainingClasses(vIndices) = vCenterClasses(i);

mNewPoints(vIndices, = round( ...

mvnrnd(mCenters(i,,cCovariance{i},nPoints));

% Count how many points are incorrectly classified

vFitClass = sign(mNewPoints(vIndices, * w - gamma * ...

ones(nPoints,1));

vZeroSpots = find(vFitClass==0);

vFitClass(vZeroSpots) = ones(length(vZeroSpots),1);

nMisclass = nMisclass + sum(vFitClass~=vCenterClasses(i));

end; %for

% Output the data points to disk

flatfile([mNewPoints vTrainingClasses],sOutputFile,1);

nTrainingClass1 = nTrainingClass1 + sum(vTrainingClasses==1);

nTrainingClassm1 = nTrainingClassm1 + sum(vTrainingClasses==-1);

end; %while

disp(sprintf('Percent separable estimate = %4.2f%%\n',100*(1-nMisclass/nTotRows)));

disp(sprintf('Number class 1 points = %d\n',nTrainingClass1));

disp(sprintf('Number class -1 points = %d\n',nTrainingClassm1));

end; %for-nDataset

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to flyingsohigh

08-31-2011 08:09 PM

You don't say what kind of help you're looking for. For general pointers, see http://blogs.sas.com/content/iml/2011/03/09/translating-a-matlab-program-into-the-sasiml-language-a-...

You can look up the various MATLAB functions at http://www.mathworks.com/help/techdoc/ to see what they do. I assume that you know what the code is supposed to be doing? If so, and assuming that you know SAS/IML, I suggest you try to translate as much as possible into SAS/IML and then ask specific SAS/IML questions. It looks like you'll need to use the following SAS/IML functions: J (for ones() and zeros(), RANDGEN (for rand()), LOC (for find()?), RandNormal (for mvnrnd()),...

For help with generating random uniform numbers, see http://blogs.sas.com/content/iml/2011/08/24/how-to-generate-random-numbers-in-sas/

For help on sampling from multivariate normal with a given covariance, see http://blogs.sas.com/content/iml/2011/01/12/sampling-from-the-multivariate-normal-distribution/