Expose, document, and demonstrate AZP variants

[jGaboardi]

Currently the standard AZP algorithm is shown in the docs and demonstrated in the Tutorials. However, there are 3 variants of AZP available in spopt:

• AZPSimulatedAnnealing
• AZPBasicTabu
• AZPReactiveTabu

These three variants are not exposed in the API and don't appear to be fully functional.

[jGaboardi]

xref – #406, #386

[jonpsy]

@jGaboardi Are we sure that the algorithms's functional logic is correct and only the API is outdated? Are there any document/benchmark for when these algorithms were running (before the API got outdated) so we can establish that logic is correct?

Details regarding this (via comment or associated links) would be helpful. Much thanks ~!

[jGaboardi]

Are we sure that the algorithms's functional logic is correct and only the API is outdated?

Nope. I have no idea. As @knaaptime pointed out here, we aren't exactly sure who contributed that. Maybe @sjsrey has a better idea?

Are there any document/benchmark for when these algorithms were running (before the API got outdated) so we can establish that logic is correct?

Another "not sure" here. At least not that I am aware of.

[knaaptime]

@jonpsy the code was ported from the outdated pysal region module, which itself was a reimplementation of clusterpy, i believe. The algorithm itself should be in good shape, afaik (i actually dont know why the api never got integrated...), but if you want to benchmark it, there's a version of AZP in pygeoda you could test against if you like. The most recent paper i'm aware of that actually uses AZP is this, but apart from that it's not seen often in the wild.

if you want to get the api in working shape, we'd gladly accept a PR :)

[jGaboardi]

we'd gladly accept a PR :)

very gladly