pandana revamp todo list

[fscottfoti]

I'm opening an issue to keep track of my to do list for the pandana rewrite taking place here

The main purpose of the rewrite is to replace the C++ api wrapper pyaccesswrap.cpp with a cython wrapper which will be much easier to use going forward. Cython was around when I worked on this but it wasn't nearly as ubiquitous as it is these days.

In addition to that I should be able to fix:

• style issues and general clunkiness in the C++ code
• use scipy NN instead of ANN, which saves us from having to build ANN
• the big one: not having global memory thus allowing Accessibility objects to be created and discarded on the fly Unable to close network in memory #52

Here's the ongoing high level todo list:

• port POI functions to cython
• port aggregation functions to cython
• port shortest path and precompute functions to cython
• integrate with @federicofernandez scipy NN code
• add unit tests for the cython wrapper (lower level tests)
• get the old unit tests back up and running again

Small issues:

• make sure Different results when making multiple calls with nearest_pois #73 is still solved
• check to see if nodes ids and locations are actually used in CH code - I don't think they are and can be removed from cython wrapper
• addGraphalg doesn't have to be public - cython shouldn't know about graphAlg
• think long and hard about which level is responsible for turning ids into indexes (CH requires indexes), in the context of node ids shouldn't have to be ints #62
• drop initialize_pois - use std::map internally, use strings for categories
• drop initialize_acc_vars - use std::map internally, use strings for categories
• pass in strings for decay and aggregation types, rather than ints
• impedance ids should be strings not ints too (leaving this as-is)
• check for memory leaks
• add test for non integer node ids
• there's probably a better way to return the POI node_ids
• fix travis build issue for python3.5

[fscottfoti]

finished with #87