Grand maybe useful?

[j6k4m8]

Hey Ben! Saw this repo and wondered if Grand meets your needs or if not, if there are improvements we could make that would get things more usable for you?

from grand import Graph
from grand.backends import DataFrameBackend
import pandas as pd

# Create an edges DataFrame
edges = pd.DataFrame({
    'source': [0, 1, 2, 3, 4],
    'target': [1, 2, 3, 4, 0],
    'weight': [1, 2, 3, 4, 5],
})

nodes = pd.DataFrame({
    'name': [0, 1, 2, 3, 4],
    'value': [1, 2, 3, 4, 5],
})

graph = Graph(backend=DataFrameBackend(edge_df=edges, node_df=nodes), directed=True)

(Should work on out-of-core pandas-like big data as well.)

It's useful for me to know what the shortcomings are!

[bdpedigo]

hey @j6k4m8 - that's awesome, thanks for letting me know! I simply didn't know that Grand had those tools, so I'll have to see if it meets the same needs or not.

My initial impression from looking at the Grand wiki is that what I'm more interested in here is a "dialect" that I haven't found in a graph library to my satisfaction, which would mimic the syntax of pandas. Not for the sake of performance (in fact it'll probably be much less optimized than a true graph library) but simply because im used to the pandas syntax for selecting and filtering data. For instance (from the readme), much like a pandas groupby:

for color, subgraph in nf.groupby_nodes("color", axis="both"):
    print(color)
    print(subgraph.edges)

gives

('blue', 'blue')
  source  target  weight
1       1       2       2
3       2       1       4
('blue', 'red')
  source  target  weight
2       2       3       3
('red', 'blue')
  source  target  weight
0       0       1       1
('red', 'red')
  source  target  weight
4       3       0       5

[j6k4m8]

Ooh neat! grand-cypher kinda gets close, though we don't fully meet the cypher spec yet so groupbys are a bit weak...

I've really liked the few minutes I've spent with the networkframe query style so far! Food for thought, wonder if there are synergies in the future :)

[bdpedigo]

Ooh neat! grand-cypher kinda gets close, though we don't fully meet the cypher spec yet so groupbys are a bit weak...

ah, cool! that looks like similar idea but for sql syntax?

I've really liked the few minutes I've spent with the networkframe query style so far! Food for thought, wonder if there are synergies in the future :)

Great! FWIW this is something I pick up and put down periodically if it's helpful for a specific project, so I'm sure there are issues and things that could be improved. But would love to look for synergies!

[j6k4m8]

very sql-like, though specifically targeting the neo4j-popularized cypher language... so everything that works in grand-cypher also works on a neo4j database.

definitely lots of good crossover! easy one that comes to mind is if we could abstract away the underlying data structures and get you access to networkframe queries in, say, graph databases and sql just as easily, through grand. thinking out loud, just starting some threads we can pick back up if it ever becomes relevant!

[bdpedigo]

that sounds great to me, the drawback of what i was doing was always that relying on the dataframes themselves is going to be suboptimal for many graph queries, so abstracting that away but keeping the syntax would be lovely