The followng were comments I recieved on my Request for Comment submitted a bit ago.
- Questions about Request for
Comment:
- What should I prioritize? NOGR or Labeled Array Interface?
- Labeled Array. This is critical to get correct, and will make NOGR need and scope clearer.
- How deep into PySAL should the Labeled Array interface go?
- Design it like the library were getting built now.
- Do not fail on import. Instead, use soft dependencies/optional import patterns
- if necessary, write Python3-only components safely, so that new features can be leveraged.
- What should get deprecated?
- Anything that looks less smooth in the unlabeled array interface should get flagged with a depwarning.
- If the tabular IO is smooth and works parallel to the older interface, then throw a deprecation warning on the FileIO components.
- Deliverables in the medium term (targeting midterm eval for GSOC):
- Two Contrib Modules:
- GeoTable: interfaces between PySAL labeled arrays & Geopandas arrays
- Pdio: extend and improve tabular interface already in PySAL
- Some work in core:
- Polymorphic weights constructors
- i.e. work on any arbitrary iterable of shapes
- return correct weights object from the iterable
- possibly indexed by a second collection of indices
- Revamp & scaffold new IO system revolving around multiple alternative
packages & their drivers:
- expose all pandas
read_functions - ensure pysal objects get serialized correctly into wkb/wkt by
to_on dataframes - wrap Fiona & geopandas constructors to provide identical output to pdio.read_files
- expose all pandas
- Plan to connect with new geopandas contributors at SciPy
- Investigate possibility of serializing with Libfeather (remote, if time remaining)
Originally posted on yetanothergeographer.tumblr.com.
gsoc