Open xiaoshijian opened 2 years ago
Hi @xiaoshijian,
Thank you for your idea! I like the way you accelerate by reducing the comparison dynamically. It is very smart. If you still have interests in improving this repository, you might want to reproduce the results and the time overhead on your training data with/without your updates. If the improvements are significant, then we can merge your solution into the main branch. Thanks!
Hello Chuanyu, thanks for introducing this method for me to test dependency I am now working on the competition for HM recommendation system in kaggle, I want to use this code for check feature importance. There are 800000 samples (x and y are array with size 800000) in training data. I don't know whether I use the code correctly or not. I found that it is too slow when I want to operate with sample size in 10 * 5 level. Therefore I read your code and think maybe slowness is iterative comparison for code in line 67 ''' for s in x: result += pr numpy.square(self._fr(s, t, x, y) - self._f(s, x)) '''
After that, I try accelerating mvtest by avoiding large amount of comparison, Here is my idea
"""
"""