[
https://issues.apache.org/jira/browse/MATH-1423?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16068729#comment-16068729
]
Phil Steitz commented on MATH-1423:
-----------------------------------
I don't think there is a bug here - more a question for the user list and maybe
documentation improvement. In order for there to be a unique solution to the
OLS parameter estimation problem, the columns of the X matrix have to be
linearly independent and there must be at least as many rows as there are
columns. if your dimensions above are (row, column), there will be no
solution because 60 observations are not sufficient to estimate a model with
2160 independent variables). It is the XX' matrix that ends up singular in
this case, which is what ultimately triggers the SingularMatrixException.
If it is 2160 rows and 60 columns, that will work as long as the columns are
linearly independent. If one of your variables is (very close to) a linear
combination of some subset of the others, you will end up with a
SingularMatrixException. Check to make sure that all of the columns are
distinct and that none is just a multiple of another.
The javadoc should advertise SME and maybe explain this or provide a link to a
reference on OLS.
> SingularMatrixException on Non-Square Matrix
> --------------------------------------------
>
> Key: MATH-1423
> URL: https://issues.apache.org/jira/browse/MATH-1423
> Project: Commons Math
> Issue Type: Bug
> Affects Versions: 3.5
> Environment: Oracle JDK 1.8.121
> Reporter: Cody Warren
> Priority: Minor
> Labels: OLSMutlipleRegression, SingularMatrixException
> Original Estimate: 1h
> Remaining Estimate: 1h
>
> I'm trying to implement an OLSMultipleLinearRegression class in the apache
> commons math java library and I keep getting a "SingularMatrixException".
> This is confusing to me because my data isn't even square (60 x 2160) which I
> thought was a requirement for a Singular Matrix.
> I've played with the data by pruning rows off and adding them back on, and
> found differing numbers of rows that will work/fail with this dataset.
> Also, I've checked my matrix for columns or rows that are full of zeros as
> suggested in this post:
> Using Apache Library for OLS Regression : Matrix is singular exception
> Is there something else with this library that I don't understand? Is there a
> way I can make this more robust or predict a singular array beforehand?
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
