[livelike.acs.puma] -- storing `puma` objects -- to/from class methods for pickle & parquet

[jGaboardi]

storing puma objects

to/from class methods

pickle & parquet

Teams idea dump

---

A idea I had yesterday that I need to put in a ticket is are to_*()/from_*() methods in {livelike,pronto_gatto}.acs.puma(). Where if the object is pickled ( puma.to_pickle(path) ) we can call acs.puma.from_pickle(path) and return the normal class instance.
The next logical step in thought is... how can we make a single parquet of a puma() class instance...

• pandas.DataFrames can store a ton of different stuff
• we can even store DFs/GDFs in cells of a DF
• so... I think...
  • we can store puma class instances as parquet'ed DFs
  • singel column
  • indexed by attribute name
  • column values are attribute values
• so the have another set of class methods to handle this
  • puma.to_parquet(path)
  • puma.from_parquet(path)

---

• for parquet specifically
  • row- column based vs. column-based...?
• structure
  • row 1 - name (str)
  • row 2 - weights (numpy.array)
  • row 3 - contraints (pandas.df)
  • row 4 - geoms (geopandas.gdf)
  • ....
• We can (theoretically) store all attributes like this -- though maybe column-based instead of row based.
• That way we would be able to store/ship around pre-built pumas in smaller format that pickle.

---

Big Qs

1. Can this parquet solution be accomplished? I think yes, but not confirmed
2. Can the parquet solution be a single column of rows vs. a single row of columns?

---

xref:

• internal livelike#-313
• internal pronto_gatto#-120
• Migrate issues from GL -- GH #9