The correct social flow normalization by source

[thekingofkings]

What is normalization by source?

We use out-flow to calculate the social lag of node i.

The intuition is that the source (home) node_i have outgoing flow to several nearby destinations (work place). These work flow will definitely go back to the node_i. This represents the influence from work place on home location.

Therefore the key is to use out flow to construct the social flow.

Previous wrong method

Previously, I always use in flow j -> i to propagate crime into node i. When normalize by source, the j -> i flow is normalized by the total out-going flow of j.

We should notice that j ->i does not equal to i -> j.

In a social science setting, the i -> j flow is more intriguing.

[thekingofkings]

To fix the normalization by source, two changes are needed to made:

• In the _generate_sociallag function of python\FeatureUtils.py.
• In the leaveOneOut function of R\NBUtils.R

[thekingofkings]

Also refer to Issue #2 to make sure the transpose is applied at the right place.

[thekingofkings]

fixed