Hello, i have a question regarding the rolX implementation:
In function make_sense() we calc matrix M, compute K and just returning it.
As far as i can see the original paper on chapter 5 (here) is doing another step before returning K: there will be calculated another K' with only one role and then returning (K / K') to get role-contribution compared to the default-contribution (K').
This is the original paragraph:
NodeSense then computes a nonnegative
matrix E such that G * E = M. The matrix E represents
the role contribution to node measurements. A default
matrix E' is also computed by using G' = ones(n, 1), where
the n nodes belong to one role. Then, for each role r and
for each measurement s, NodeSense computes E(r,s) / E'(r,s).
Note: Matrix K in the code coressponds to matrix E in the paper
Is this step missing in the code or should this step be added to be accurate to the original paper?
Thanks in advance
https://github.com/Lab41/Circulo/blob/77692ff21566a721d4bf45c0d88053f9cf2bfa93/circulo/algorithms/rolx.py#L310-L313
Hello, i have a question regarding the rolX implementation: In function
make_sense()
we calc matrix M, compute K and just returning it. As far as i can see the original paper on chapter 5 (here) is doing another step before returning K: there will be calculated another K' with only one role and then returning (K / K') to get role-contribution compared to the default-contribution (K').This is the original paragraph:
Is this step missing in the code or should this step be added to be accurate to the original paper? Thanks in advance