Broken "sort groups by..." for multiple sort criteria

[thinkh]

• Release number or git hash: 169bd48
• Web browser version and OS: Chrome 75

Steps to reproduce

I could not provide a sandbox link, as a sorting bug was fixed with bafe482 (after v4.0.0-alpha.11) and no alpha.12 is available (yet).

1. Run npm start and open http://localhost:8080/aggregate.html
2. Column cat -> column menu -> sort groups by... -> descending
3. Column cat2 -> column menu -> sort groups by... -> descending/ascending

Observed behavior

Only the first group sorting is applied (in the example cat) and the second group sorting has no effect.

Expected behavior

• All group sort criteria should be considered

[sgratzl]

so far when group sorting by a categorical value it actually sorts by two criteria: the name of the most frequent group, and second its count.

lineupjs/src/model/internalCategorical.ts

Lines 101 to 110 in 1019f08

	export function toGroupCompareCategoryValue(rows: ISequence<IDataRow>, col: ICategoricalColumn, valueCache?: ISequence<ICategory | null>): ICompareValue[] {
	if (isSeqEmpty(rows)) {
	return [NaN, 0];
	}
	const mostFrequent = findMostFrequent(rows.map((d) => col.getCategory(d)), valueCache);
	if (mostFrequent.cat == null) {
	return [NaN, 0];
	}
	return [mostFrequent.cat.value, mostFrequent.count];
	}

e.g. in builder3.html

first criteria level: Cat, 2nd. Cat2

values by group:

1. c1,c2:
2. most frequent c1, count 8
3. most frequent c2, count 8
4. c1,c3:
5. most frequent c1, count 8
6. most frequent c2, count 8
7. c1, c1
8. most frequent c1, 12
9. most frequent c1, 12

sorting asc it means: c1,c2 < c1,c1 since the group since of the first group sort criteria is smaller.

[thinkh]

Thanks for the explanation.

What does most frequent group mean in this case? Is it the size (number of items) of a group?

I think for the user it's very opaque what the group sorting criteria is. In my opinion we have the following options:

1. Sort groups by category name only: Pretty obvious, since it works the same as the item sorting hierarchy and the category name is what the user can see
   
2. Sort groups by most frequent group: Only visible to the user if the group name column is available. It needs to better communicated/displayed to the user.
3. User choice: Alternatively, we can also leave the choice to the user and add another radio button to the sort groups by dialog with the options category title (default) and category size. It could look similar to the group by dialog of a string column:
   

@sgratzl What do you think?

[thinkh]

After checking issue #109 it seems like these two issues are pretty similar.

@sgratzl You posted already a screenshot #109 (comment) which matches my feature request in the previous comment.

Maybe we move the discussion to issue #109 and continue there. I still owe you a comment there. 😃

[sgratzl]

In the end a simple compromise is to sort groups of categoricals by their most frequent category name only. Thus, in case it is used as a grouping criteria it behaves as expected when combined with another group sorting criteria