search

czbiohub-sf / xicor

xi correlation method adapted for python
MIT License
145 stars 17 forks source link

Result depends on arguments order #22

Open slavnycoder opened 2 years ago

slavnycoder commented 2 years ago

xi(x, y) is not the same as xi(y, x) how to decide which order should i use?

stfgrm2 commented 2 years ago

As per the paper describing this new measure, it should not be. xi(x,y) tells you how much y is a function of x. As not all functions are invertible, this is not the same as xi(y,x), which is how much x is a function of y. If you're trying to predict y from x, use xi(x,y). If you want some symmetric measure, the paper suggests to take max(xi(x,y), xi(y,x))