OWCorrelations: Use heuristic to get the most promising attribute pairs

[VesnaT]

When dataset is too big (i.e too many attributes or instances), it is impossible to calculate correlations for all attribute pairs. KMeans is used to suggest the most promising attribute pairs by selecting the pair within a cluster.

[BlazZupan]

Please sort the rows by absolute value of correlations. Also, keep the same precision when displaying numbers (e.g., instead of 0.76 write 0.760).

[BlazZupan]

Please show progress bar.

[VesnaT]

Do not merge until
biolab/orange3#2117 and
biolab/orange3#2118
are merged.

[ajdapretnar]

Since both necessary commits are merged, can we merge this one as well? @BlazZupan

[astaric]

@BlazZupan, this has been sitting here for more than a year. Is there still need for this widget? Do we ask @VesnaT to rebase, or do we close it as won't fix?

[ajdapretnar]

@VesnaT Could you please rebase so we can merge and hopefully include it in the migration of the widget? Thanks!

[BlazZupan]

It would be great if we can have this widget finally in Orange.

[VesnaT]

It's rebased, but the tests don't pass (I guess since vizrank has been changed). The tests should be fixed before the widget is moved to Orange.

[ajdapretnar]

Yeah, it's probably #155.

[VesnaT]

The last commit fixes #155

[ajdapretnar]

I have tried this and while it looks really good, I am getting a segfault when switching between Pearson and Spearman. I cannot easily replicate this, but it happens each and every time when using iris.tab, clicking on different feature pairs and switching between Pearson and Spearman. Normally I select a feature pair and then switch the method.

[ajdapretnar]

I merged this. I will have a look at the segfault issue and try to have it as replicable as possible. This will be a separate issue.