[graph-tool] GraphView with lambda function

Tiago de Paula Peixoto tiago at skewed.de
Mon Jan 6 15:11:53 CET 2020


Am 06.01.20 um 13:53 schrieb Gerion Entrup:
> Hi,
> 
> the documentation describes that GraphView can be given a unary function
> to filter vertices or edges.
> 
> I have tried that and it seems to fail. My GraphView has not the
> expected vertices and edges.
> 
> However, my assumption is that the filter is only evaluated one time
> (at initialization).
> 
> Let me make an example:
> 
> ```
> g = graph_tool.Graph()
> a = graph_tool.GraphView(g, vfilt=...)
> 
> fill_the_graph(g)
> do_stuff_with(a) # <- here a does not contain any data
> ```
> 
> From the documentation, it seems that graph_tool constructs a property
> from the filter function and uses this for filtering (therefore also
> needing O(N)), but fill this property only on construction. Can you
> mention this in the documentation as a hint or warning?

Right, this is entirely expected behavior. It seems obvious to me in the
documentation, but I will make it more explicit.

Note that it would be completely unreasonable performance-wise to
populate the filter property map lazily on demand.

Note also that if you had modified `a` instead of `g` in your example,
the filtering would behave as expected (i.e. new vertices or edges would
appear in the graph view).

> Maybe also a recalculate function for GraphView is meaningful that
> evaluates the lambda function again.

I don't think this is good design. GraphViews are supposed to be cheap
objects that can be constructed on demand. If the filtering needs to be
re-done, then a new GraphView should be constructed, maybe even composed
from the older one. I.e. in your example you would re-create `a` after
you had modified `g`.

Best,
Tiago

-- 
Tiago de Paula Peixoto <tiago at skewed.de>


More information about the graph-tool mailing list