search

Xtra-Computing / thundersvm

ThunderSVM: A Fast SVM Library on GPUs and CPUs
Apache License 2.0
1.56k stars 217 forks source link

Unresponsiveness to interrupts on Google Colaboratory #208

Open ArktinenSieni opened 4 years ago

ArktinenSieni commented 4 years ago

Hey! I'm running the library on Google Colaboratory. I am running to issues where fitting of a classifier hangs. Sometimes the same data-set runs in few minutes, and rest of the times the fitting is left hanging. I am not sure if this is due to my hyperparameters, or the Colaboratory's resource management.

My issue is that the Google Colaboratory is unable to interrupt the fitting process. Instead of interruption of single notebook-cell, I have to reset whole notebook to retry the process. In this manner I lose all my variables and progress located in the local memory.

Would there be a way to either interrupt the process safely, or set a time-out to the fitting?

zeyiwen commented 4 years ago

There is no easy way to do in the current version. You can change the maximum number of iterations in ThunderSVM in C++, and recompile the library.