Documenting behavior for aggregations that can't be calculated about pandana

[smmaurer]

Related to this are the messages about dropped rows that sometimes show up when you run an aggregation calculation:

Computing pop_500_walk
Removed 189769 rows because they contain missing values

These messages are generated by the pandana.Network.set() call that links the values being aggregated to the network.

https://github.com/UDST/pandana/blob/master/pandana/network.py#L235

Here's what happens, for the example of aggregating a variable from the households table:

• if filters are provided, these rows are removed from the households table first
• then, pandana tries to link each remaining row to a network node
• any rows that are either missing a node id or have a NaN in the column being aggregated are dropped
• the message refers to rows dropped from the households table, not nodes dropped from the network

Often, the rows are dropped because they can't be matched to nodes (for example households that are not assigned to buildings and thus don't have a spatial location), not because of missing values in the data column.

Rows that are explicitly filtered out aren't counted, which can result in variations in the number of rows dropped for aggregations in the same table.

Here is a notebook where we dug into this: More-aggregation-troubleshooting.ipynb