# Regression with tall array (Using datastore, CSV) - Error

3 views (last 30 days)
K.P. on 12 Jul 2021
Edited: K.P. on 1 Aug 2021
Hi
dpb on 12 Jul 2021
I just tried to see if it was tall arrays and fitglm
>> X=[1:1000].'; X=tall(X);
>> Y=randn(size(X)); % this is interesting sidelight on the way...
Error using randn
Size inputs must be numeric.
>> size(X)
ans =
1×2 tall double row vector
1000 1
>> Y=randn(1000,1); Y=tall(Y); % OK, have to brute-force it
>> fitglm(X,Y,'Distribution',"normal")
Iteration : 0% completed
Iteration : 50% completed
Iteration : 100% completed
Iteration : 0% completed
Iteration : 50% completed
Iteration : 100% completed
Iteration : 0% completed
Iteration : 100% completed
ans =
Compact generalized linear regression model:
y ~ 1 + x1
Distribution = Normal
Estimated Coefficients:
Estimate SE tStat pValue
__________ __________ ________ _______
(Intercept) 0.0015036 0.064429 0.023338 0.98139
x1 1.6177e-05 0.00011151 0.14507 0.88468
1000 observations, 998 error degrees of freedom
Estimated Dispersion: 1.04
F-statistic vs. constant model: 0.021, p-value = 0.885
>>
So, fitglm will accept tall arrays; the syntax must be else where it would seem...

Ive J on 13 Jul 2021
Edited: Ive J on 13 Jul 2021
Well, your data is tall table, and that's what MATLAB complains about: since your first argument is a table, MATLAB thinks y is modelspec. You have two options:
% 1-feed fitglm with matrix
mdl = fitglm(x{:, :}, y{:, :}, 'Link', 'logit', 'Distribution', 'binomial');
% 2-OR: merge x and y as a table
data = [x, y]; % last column is the dependent variable by default
mdl = fitglm(data, 'Link', 'logit', 'Distribution', 'binomial');
Btw, your data is fairly small and (I assume) fits within memory, tall arrays should be avoided for such small datasets.
##### 2 CommentsShowHide 1 older comment
Ive J on 13 Jul 2021
If I were you I would also test with arrays. Processing tables is almost always (based on my experience) slower than arrays.
Good luck!