[WIP] Adding pure SQL GPU-BDB Queries

[DaceT]

Fixes #230

All of the queries (01, 06, 07, 09 11, 12,13,14,15,16,17,20,21, 23, 24, 29) run successfully except for q22. However, once that is up and running, I'll commit the changes here.

[VibhuJawa]

Thanks for working on this @DaceT .

I dont think we should rely on os.envron when we are running these queries. We should read it via a config file and then check based on the dataframe passed.

A big problem with it is the reduced readabilty of import pandas as cudf means that when a trace comes it will be insanely hard to reason about something that says its coming from cuDF but coming from pandas and vice versa.

Also, environment variable based optional imports look fidgety to me , especially with things that get bundled up as a package and then dask uses them. I think it can cause a lot of problems downstream.

[VibhuJawa]

I would have us pass a ** kwarg and expand this class .

gpu-bdb/gpu_bdb/bdb_tools/readers.py

Lines 90 to 97 in db207a1

	def __init__(
	self, basepath, split_row_groups=False,
	):
	self.table_path_mapping = {
	table: os.path.join(basepath, table, "*.parquet") for table in TABLE_NAMES
	}
	self.split_row_groups = split_row_groups

So something like:

 def __init__( self, basepath, split_row_groups=False,  backend = 'GPU'): 
     self.table_path_mapping = { 
         table: os.path.join(basepath, table, "*.parquet") for table in TABLE_NAMES 
     } 
     self.split_row_groups = split_row_groups 
    if back_end =='CPU'
      self.back_end = dask_cudf
   else:
      self.back_end = dask.dataframe

And Call it like :

self.back_end.read_parquet

[DaceT]

Wouldn't we want the opposite or did you mean to put 'GPU' instead of 'CPU'? @VibhuJawa

if back_end =='CPU'
      self.back_end = dask.dataframe
else:
      self.back_end = dask_cudf

[VibhuJawa]

yup. Exactly that.

[VibhuJawa]

Just use to_frame below.

    sales_corr = result["x"].corr(result["y"]).compute()
    result_df = sales_corr.to_frame()

[VibhuJawa]

LGTM !

One small style fix but everything else looks good.

[VibhuJawa]

LGTM