Getting negative entropy/positive Log likelihood for a weighted network with positive edge weights.
Dear Tiago, I am trying to model a weighted dense network with edge weights in the range 0 to 1, with the non-hierarchical SBM for both degree corrected and non-degree corrected versions. However, I get negative entropy for both dc and ndc SBMs. I have attached a sample network for your consideration along with a minimal working example from my code. Thanks for your time! Sukrit sukrit_network.gz <http://main-discussion-list-for-the-graph-tool-project.982480.n3.nabble.com/file/t496012/sukrit_network.gz> working_example_W_SBM.py <http://main-discussion-list-for-the-graph-tool-project.982480.n3.nabble.com/file/t496012/working_example_W_SBM.py> -- Sent from: http://main-discussion-list-for-the-graph-tool-project.982480.n3.nabble.com/
Am 10.01.19 um 04:27 schrieb isukritgupta:
Dear Tiago, I am trying to model a weighted dense network with edge weights in the range 0 to 1, with the non-hierarchical SBM for both degree corrected and non-degree corrected versions. However, I get negative entropy for both dc and ndc SBMs. I have attached a sample network for your consideration along with a minimal working example from my code. Thanks for your time!
When using real edge covariates, the overall likelihood becomes a probability *density*. Since the probability density can exceed 1 in value, its log can be positive, and hence the entropy can be negative. This is normal. -- Tiago de Paula Peixoto <tiago@skewed.de>
participants (2)
-
isukritgupta -
Tiago de Paula Peixoto