using a setup function in benchmarks

[akshitasure12]

I expect that we will need many different types of graphs for the algorithms. So the get_graph May not be able to be used very often. At least that is what happens with tests and pytest. We create a test class and put some graphs that could be used by lots of tests, but then they end up only being used by a couple of the tests and the rest construct their own graph. So we may want to create a different setup function for many of the timing functions. I think we’ll have to wait to see what is most useful. Keeping it as a common function is fine with me. Could we name it something like: cached_gnp_random_graph?

Originally posted by @dschult in #24 (comment)

I was working through the benchmarks and realized that for the tournament functions, we’re currently including the graph generation time in the benchmark measurements, possibly skewing the results.

def time_is_reachable(self, backend, num_nodes):
        G = nx.tournament.random_tournament(num_nodes, seed=42)
        _ = nx.tournament.is_reachable(G, 0, num_nodes - 1, backend=backend)

Following the way benchmarks are implemented in NetworkX, I was thinking of introducing a setup function in nx-parallel, at least for the tournament benchmark, which seems to need a different graph structure. That said, I was also wondering if we should aim for a more uniform setup approach across all benchmarks.

[Note]: I read through PR#24, and it looks like most benchmarks work well with the current cached graph approach. If we decide to continue using the cached graph, perhaps we could introduce a dedicated setup function only for the tournament benchmark. Please let me know your thoughts on this.

[dschult]

Yes -- I am used to a setup function being used with asv for each timing script. Sometimes you can use the same setup for multiple functions, but I've often seen a different setup for each timing.

In this case the tournament graphs are special so should have a setup to avoid building the graph inside the function. And we do not want to use any kind of centralized setup function. At least as far as I can imagine.

[akshitasure12]

I will raise a PR for the same, thank you :)