[ENH] OWRandomize: Add a new widget

[VesnaT]

Issue

Create a new widget for shuffling columns in a data able.

Description of changes

Enable multiple types of randomization at the same time.
New widget Randomize.

Includes

• Code changes
• Tests
• Documentation

[codecov-io]

Current coverage is 89.28% (diff: 100%)

Merging #1863 into master will increase coverage by <.01%

@@             master      #1863   diff @@
==========================================
  Files            86         86          
  Lines          9116       9115     -1   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
  Hits           8138       8138          
+ Misses          978        977     -1   
  Partials          0          0          

Powered by Codecov. Last update 463242d...3052123

[janezd]

type_ instead of _type (like exec_ in main. Where, by the way, the underscore is no longer necessary in Python 3, AFAIK).

[janezd]

Use lowercase here (Shuffle: Classes, Metas doesn't look nice)

[janezd]

I prefer printing "a, b, c, d and e", where possible. Try something like

def send_report(self):
    labels = ["classes", "features", "metas"]
    include = [self.shuffle_class, self.shuffle_attrs, self.shuffle_metas]
    include = [label for labels, i in zip(labels, include)
    text = " and ".join(filter(None, (", ".join(include[:-1]), include[-1:]))) or "none"
    self.report_items ...

This doesn't require if. If you find the last line too cryptic, you can get rid of filter by handling len(include) == 1 separately.

[janezd]

I'd prefer text = "none".

[janezd]

Vertical arrangement of check boxes with such short labels doesn't look nice. I think this is nicer:

    box = gui.hBox(self.controlArea, "Shuffled columns")
    box.layout().setSpacing(20)

That is, hBox, longer (more informative) title, and spacing.

[janezd]

If the box title is changed, perhaps also change "Shuffle" to "Shuffled columns".

[janezd]

"of rows" would be more accurate than "of data". However, you can change "Scope" is changed to "Proportion of shuffled rows", and then put just "{} %".format(self.scope_prop).

[janezd]

This is now no longer an enum. Why not something like this?

Type = Enum("Randomize", dict(RandomizeClasses=1, RandomizeAttributes=2, RandomizeMetas=4))

[VesnaT]

Why should it be an enum? It is no longer enum, because they don't support bitwise operators.
If you insist with enums, I should probably add two new parameters to the method (for each type of randomization). Or am I missing something here?

[kernc]

Enums accept a type parameter, e.g. Enum(..., type=int).

[VesnaT]

Thanks!

[janezd]

Otherwise, no, it doesn't have to be Enum. We're mostly not using it. I just liked it and asked why it was removed.

[janezd]

I'd prefer True and False, so everybody knows that these are boolean flags, not numbers. The trick in apply will still work.

[janezd]

This is exactly how I'd do it and I like it. But perhaps

type_ = sum(part for part, flag in zip(Randomize.Type, [self.shuffle_class, shuffle_attrs, shuffle_metas]) if flag)

is better. (If you use [self.shuffle_class, shuffle_attrs, shuffle_metas] in a lot of places, you can create a property

@property
 def parts(self):
     return [self.shuffle_class, shuffle_attrs, shuffle_metas]

and simplify this expression and also send_report.

[janezd]

I'd prefer a gui like this.

[VesnaT]

Done