kopt | Hyper-parameter optimization for Keras | Machine Learning library

My jobs have been suffering due to segmentation faults when calling glmnet (downloaded from here:http://web.stanford.edu/~hastie/glmnet_matlab/download.html) from my MATLAB code. I call the glmnet routine thousands of times. I have noticed the following peculiarities about the problem occurence:

1. The problem is more frequent when the size of my input matrices are larger.
2. I use both gaussian and poisson distribution in separate jobs, and I notice that the problem is more frequent when fitting the Poisson distribution (which also takes usually longer to converge, so might involve more loops internally?) Since there haven't been reports of segmentation faults for the R version for these two distributions, my suspicion is that the problem, likely a memory leak, might lie in the mex interface rather than the core glmnet Fortran code, which I am pasting below. Any insights into where a memory leak might be happening is greatly appreciated! Apologies for the lengthy code dump.

Thanks!