Re: [julia-dev] Scalars and 1x1 arrays

[Stefan Karpinski]

That's correct, 1x1 matrices and scalars are different. Multiplying a matrix by a scalar is fine — it scales the matrix. Multiplying a matrix by a 1x1 matrix is an error unless the matrix has a corresponding dimension of size 1.

On Aug 16, 2012, at 12:03 AM, Tirthankar Chakravarty <tirthank...@gmail.com> wrote:

I am trying to code up some examples in Julia and have some questions regarding arrays and scalars. Here is Matlab code that works:

%==============================

clc; clear;

vY = randn(100, 1);  % outcome variable

mX = randn(100, 4);  % design matrix

iN = size(vY, 1); % sample size

vBeta = (vY\mX)';  % estimated coefficients

vE = vY - mX*vBeta;  % residuals

dSigmaSq = vE'*vE/iN;  % residual variance

mV = dSigmaSq.*(inv(mX'*mX)); % covariance matrix 

vStdErr = diag(mV);  % std. err. 

vT = (sqrt(iN)*vBeta)./vStdErr;  % t-statistics

[vBeta, vStdErr, vT]

%==============================

On line 7, Matlab understands that the object "dSigmaSq" is a real scalar.

However, Julia needs this value to be explicitly dealt out of the 1x1 array,

#===============================

vY = randn(100, 1);  # outcome variable

mX = randn(100, 4);  # design matrix

iN = size(vY, 1); # sample size

vBeta = (vY\mX)';  # estimated coefficients

vE = vY - mX*vBeta;  # residuals

dSigmaSq = vE'*vE/iN;  # residual variance

mV = dSigmaSq[1,1].*(inv(mX'*mX));  # covariance matrix; dSigmaSq.* 

vStdErr = diag(mV)  # std. err.

vT = (sqrt(iN).*vBeta)./vStdErr  # t-statistics

println([vBeta'; vStdErr'; vT']')

#===============================

Is this an intended design feature? I am wondering if my expectations are skewed by Matlab's behaviour.

Thanks.

--