Moving towards:
* A netdis function that takes two input graphs and a reference graph.
* Sub-functions that follow through paper steps, sequentially applied with output of one providing input for next.

Code to filter ego networks with fewer than threshold nodes/edges was duplicated many places. Added that functionality to make_named_ego_graph, where ego networks originally generated.

Also, remove unnecessary duplication of simplifying in read_simple_graphs function, as it's already performed in read_simple_graph (which read_simple_graphs calls).

Default behaviour of ego network generating functions is now min_ego_nodes <- 3 and min_ego_edges <- 1. Set these to zero in function calls so orca tests now passing.

Should implement tests to check that filtering of small ego networks is working as expected (i.e. when min_ego_edges and min_ego_nodes > 0)

ego network generating functions now take min_ego_edges and min_ego_nodes arguments to filter out small ego networks, which default to 1 and 3 respectively. Set to zero for tests.

Filtering is probably happening explicitly elsewhere in the tests - could be removed with min_ego_edges and min_ego_nodes given non-zero values as appropriate.

Vignette to calculate netdis for 2 query graphs and a reference network:
* Follows steps of paper.
* Output of previous step used as input to next step (as much as possible).
* Calculates netdis for all graphlet sizes up to max_graphlet_size
* No explicit calls to purrr - all hidden in functions for each step.
* Needs a lot of tidying of function names, docs, removal of potentially obsolete functions etc.

Questions:

* What should be in top level netdis function - current netdis function takes centred graphlet counts (i.e. output of all but last step). Should that be parent function taking two query graphs, ref graph, and parameters?

* How to implement comparison of many graphs? This may differ depending on choice above - e.g. when comparing many graphs ideally don't want to be re-calculating ego nets and graphlet counts each time.

but bugs -
netdis_expected_graphlet_counts applies scaling so shouldn't be used for GP approx or constant count subtraction as written

+ started to implement possibility to pass in graphlet counts and a constant expected counts value

And added test for geometric poisson

…-institute/network-comparison into fastSmoothEMD-review

…ce-refactor

Refactoring of Netdis code and linting of whole package (including NetEMD).

- Luis has reviewed the code and approved it being merged.
- Andrew is also ok for NetEMD linting changes to be merged.

Creation of a the simple case usage of the different network comparison methods with their default settings.

…t_emd function to net_emd_one_to_one. It now accepts graph objects too.