Node equivalence about hatchet HOT 4 CLOSED

Hey @slabasan:

This is really a design question — we need both behaviors but the question is where they should happen

Since the nodes are the thing that ultimately goes in the DataFrame in the index, I think that id(self) == id(other) (aka self is other) should be the behavior we want there. Every DF should have nodes from only one graph, and in that scenario, comparing by node id makes sense. For trees that is effectively the same as comparing by callpath (only faster, since comparing ids is faster than comparing tuples). For graphs it’s not general enough -- you need to build the graph and compare nodes to determine which nodes are unique, since not every graph node has a unique callpath as in a CCT.

This last bit is key -- there shouldn’t be a callpath on the Node in the final version of this, specifically because comparing the callpath isn’t general enough for graphs. We have to build the graph and use node identity (is, or id(self) == id(other)) for graphs. So, before you can compare graph nodes in hatchet, you should unify the graphs, and you should traverse and compare frame values to do that.

There are probably other situations (besides unification) where comparing frame values between nodes from separate graphs makes sense, but i would stick to always unify for now.

from hatchet.

Thanks @tgamblin for clarifying. So if we have a unify in graphframe, let's say it returns two graphframes with updated graphs and dataframes. If we are trying to unify two graphframes (i.e., identical input data set), the unify should update the node ID's so they are consistent between both, correct? Then we should also save the mapping from the old nodes to the new nodes, which can be used to update the dataframe.

from hatchet.

@slabasan: I'd implement union on Graph first, and have it return two dicts:

def union(left, right):
    """Union two Graphs and return the resulting graph along with mappings
       from the old graphs' nodes to the new graph's nodes.

    Returns:
        (Graph, dict, dict): unioned Graph and mappings from left and right ids (respectively)
             to Graph's ids.
    """

then implement unify on GraphFrame and have it call union and use the mappings to translate. I think this is basically what you're suggesting, right?

from hatchet.

I think this issue has been resolved. The node equivalence check is correct behavior, and we are working on implementing the union operation on a graph.

from hatchet.