Very long training time for Single Feature SVM (while running a greedy feature selection scan) #5
Description
I'm running a greedy feature selection scan, the first step is finding the first best feature by training a classifier for each feature and choosing the best one. then scanning for the next etc.
i have ~170 features and onc i've chosen the first feature the process is satisfactorily fast (few seconds for each training).
my problem is the first step. for some weird reason, training with just one feature takes AGES. more than 40 minutes for a single training. at this rate it'll take a few days to select the first feature and then a couple of hours for the rest. that's crazy.
i'm considering trying something else for the first step, as it is actually a threshold problem... but still, this shouldn't be happening.