Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

** On entry to DLASCL parameter number 4 had an illegal value Crashes Python #3596

Closed
Berowne opened this issue Apr 6, 2017 · 8 comments
Closed

Comments

@Berowne
Copy link

Berowne commented Apr 6, 2017

G'day.

I'm trying to test the VAR model on multivariate TS.
http://statsmodels.sourceforge.net/devel/vector_ar.html#granger-causality
from statsmodels.tsa.vector_ar.var_model import VAR

when I run
results=model.fit(2)
I get the error
** On entry to DLASCL parameter number 4 had an illegal value

@bashtage
Copy link
Member

bashtage commented Apr 6, 2017

There is a nan or inf in the matrix being inverted.

@Berowne
Copy link
Author

Berowne commented Apr 6, 2017 via email

@bashtage
Copy link
Member

bashtage commented Apr 7, 2017

This crash is a feature of Intel's MKL, not Python/statsmdodels/numpy. Best would be to check for nan/inf in sm and raise with a nice error message.

@bashtage
Copy link
Member

bashtage commented Jun 8, 2019

Closing since fixed as far as it can since mostly a MKL issue.

@dario-passos
Copy link

I can reproduce this error as well! But my fist notice of a similar error came from sklearn PLSRegression model fit. It raises a problem with linalg svd decomposition but only for the first time it is run. If I run the same cell for a second time it will work, Any clues from where this problem might come in your case with the ARIMA model?

Cheers
Dário

@bashtage
Copy link
Member

bashtage commented Aug 11, 2020

@dario-passos Are you using Windows by chance?

If you are:

  1. Using Windows version 2004 or later (build 19041 or later)
  2. Using NumPy installed from pip

then you might be experiencing numpy/numpy#16744

@dario-passos
Copy link

dario-passos commented Aug 11, 2020

@bashtage , yes, I'm on a Windows 10 (version 2004) PC and I have my numpy installed through pip!
A few minutes ago I installed the numpy+mkl version on my system and those svd errors seem to have disappeared. This might be in line with the OpenBlas issues that the post you mention talk about. However, for me, this solution was not the best one because my tensorflow-gpu 2.2 install is now giving me warnings... But that is a battle to another day. Thanks for sharing the information. I'll definitely take a look.
Cheers
Dário

@bashtage
Copy link
Member

This OpenBLAS bug is definitely a nasty one that will be catching a lot out until fixed.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants