estimate_betweenness() should be deprecated

[szhorvat]

TODOS

• Adapt docs in igraph.
• Make PRs in repos using estimate_betweenness() and the dotted version.
  • Replace igraph::estimate_betweenness() with igraph::betweenness() bec… SachaEpskamp/qgraph#77
  • rm igraph::estimate_ functions that will soon be deprecated thomasp85/tidygraph#174
  • replace igraph::estimate_closeness() with igraph::closeness() because… stocnet/migraph#271
  • Replace igraph::estimate_betweenness() with igraph::betweenness() because of upcoming deprecation biometry/bipartite#15
• Make PRs in repos using estimate_edge_betweenness() and the dotted version.
  • Replace edge.betweenness.estimate() with edge_betweenness() because of upcoming deprecations FMestre1/gDefrag#2
  • tidygraph see above
• Make PRs in repos using estimate_closeness() and the dotted version.
  • migraph see above

estimate_betweenness() should be deprecated and should not appear in the docs. "Estimate" is a bit of a misnomer here as proper betweenness estimation takes a lot more than what this function does. The replacement is using the cutoff parameter of betweenness().

The same applies to estimate_closeness(), but that function already doesn't appear in the docs.

Do you have time for this @maelle ?

[maelle]

What about estimate_edge_betweenness()?

[szhorvat]

The same. All the estimate_xyz() functions are replaced by a new cutoff argument in xyz(). The reason is that limiting shortest paths to a cutoff distance is not nearly enough for getting a proper estimate of the result. What these functions produce is very useful, but it's not an estimate of the cutoff-less result.

I consider this very important because many people come to igraph without much experience in network analysis. They see the name "estimate" and decide to just use this function for a faster approximate calculation. That's not what they really do. Let's avoid names that create confusion.

The "estimate" names have already been removed from the C core.

[maelle]

Ok! We'll have to start with a simple warning though, then an error in another version of the package.

[maelle]

actually, we'll have to start with a soft deprecation because some revdeps use estimate_betweenness(): https://github.com/search?q=estimate_betweenness+org%3Acran&type=code

[maelle]

I'll open PRs in the corresponding repos (not the cran/ ones, which are mirrors), hopefully next week.

[maelle]

Even edge.betweenness.estimate is used in gDefrag

[maelle]

I added a TODO list in the first comment.

[maelle]

I'll resume the PR openings next week.

[maelle]

I've opened PRs in all revdeps where I could, via a GitHub search, find usage of the to-be-deprecated functions.
#852 is ready for review.

[maelle]

The PRs have been merged/integrated into revdeps but we'd need to monitor when the updated versions go to CRAN I guess.

[maelle]

All PRs in revdeps merged/integrated except FMestre1/gDefrag#2 (comment)

[maelle]

even that one is merged now

[maelle]

I'll close this: since we use lifecycle for this deprecation, it'll be looked at when we increase deprecation levels anyway.