math domain error about causal-learn

[kunwuz]

Thanks for reaching out. We believe this issue is related to #29. The reason seems to be the violation of the required assumption of the input data, although we are not totally sure yet. Specifically, the covariance matrix might be singular so it doesn't have a proper inverse.

[asha24choudhary]

Thank you for your reply. I have the most version of the library installed where there is already a control statement to check for the singularity within the data. I am not getting what could be the reason for the negative value in the sqrt function which I described before. My data looks like this

.

And this is how I generated it

Here is the correlation matrix and the inverse covariance matrix
sub_corr_matrix = [[1. 0.98101581 0.91548873 0.96978605]
[0.98101581 1. 0.93355678 0.98870011]
[0.91548873 0.93355678 1. 0.97673877]
[0.96978605 0.98870011 0.97673877 1. ]]

inv = [[ 2.65907580e+01 -3.42492672e+01 -5.90995682e+00 1.42426984e+01]
[-3.46799908e+01 -1.83554478e+15 -1.28320417e+15 3.06815859e+15]
[-5.88988503e+00 -1.28320417e+15 -8.97070431e+14 2.14490757e+15]
[ 1.42536435e+01 3.06815859e+15 2.14490757e+15 -5.12850312e+15]]

Please help me

[asha24choudhary]

I just added some noise to the y variable. I did not get any errors. Don't know what should I conclude from this? Does this mean the input matrix should have full rank? So does this library creates a problem for the deterministic causal scenarios?

[kunwuz]

Hi, thanks for checking this. I calculated the determinant of the correlation matrix you provided and the result was close to zero. There may be some issues regarding the singularity check, or the math domain error could be due to other reasons. Let us check on this.

[jdramsey]

Just a thought, here's the Wikipedia page for partial correlation that spells out the assumptions etc.: https://en.wikipedia.org/wiki/Partial_correlation This is the "matrix inversion" method. Maybe an assumption is being violated.

…

On Wed, Jun 28, 2023 at 9:33 AM Yujia Zheng ***@***.***> wrote: Hi, thanks for checking this. I calculated the determinant of the correlation matrix you provided and the result was close to zero. There may be some issues regarding the singularity check, or the math domain error could be due to other reasons. Let us check on this. — Reply to this email directly, view it on GitHub <#119 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACLFSR2F2DHFBMPW4VWLULDXNQXCJANCNFSM6AAAAAAZWZ76ZA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[asha24choudhary]

Thank you so much for the immediate help.