I am trying to estimate the sparse inverse covariance matrix of my gaussian graphical model. I installed the glasso package in R and tried out some examples.
After that I ran the glasso software on my own data. So I fed it my empirical covariance matrix. However, it seems to get stuck and doesn't give me the results. So I was wondering if I was running the software incorrectly.
I followed this manual: http://cran.r-project.org/web/packages/glasso/glasso.pdf
So if S is my empirical covariance matrix. I just ran the following:
a <- glasso(S, rho=.01, trace=TRUE) # outer loop, m = 1 # outer loop, m = 2 After the above, it gets stuck.
I am attaching my empirical covariance matrix as well which I feed to the software.
http://rapidshare.com/files/141198734/emp_covariance_matrix.mat
Any guidance will be much appreciated.
Best Answer
Did you standardized data before estimation of covariance matrix? Standardization is necessary before you run glasso, though there may be exceptions ( however, I am not sure about those ocassions). Other option may be estimate covarience matrix and then transform it into correlation matrix. Then run glasso and see the result. Hope it works.
Similar Posts:
- Solved – R – glasso very slow for high feature space
- Solved – How to perform Cross-Validation for glasso to select lambda in R
- Solved – Graphical lasso numerical problem (not SPD matrix result)
- Solved – Graphical lasso numerical problem (not SPD matrix result)
- Solved – Empirical Likelihood estimation in R