Allow same column to be selected multiple times in y-axis selections

[mhkeller]

Feature Request

Description of Problem:

Currently when doing aggregate functions, I can't compare different aggregate functions on the same column such as mean and median.

Potential Solutions:

This would be remedied if I could add the same column to the Y Axis selection multiple times.

For example, using this data of student exams, I want to group by exam_subject and create a bar chart showing avg and median exam_score in each group. Once I select exam_score once, though, I can't select it again.

[texodus]

You can duplicate a column in Perspective by giving it an alias:

viewer.restore({
    columns: ["My Column", "My Column 2"],
    expressions: {
        "My Column 2": "\"My Column\""
    }
});

We're going to add quick-duplicate action buttons to the columns settings bar at some point, but they will just perform this action.

(Moving to discussions)

[mhkeller]

Got it. From a user perspective, some questions that come to mind are a) how to avoid having a bunch of differently named columns that are actually the same column and b) how to clean these up quickly once the user doesn't need them anymore. One UX flow could be:

1. User has avg exam_score in the Y-bar and they want median as well
2. User clicks a button or right-click's on exam_score box to duplicate it and the second exam_score auto-populates in the Y-bar area with the same name exam_score or a small HTML label that says it's a copy. It would be nice to avoid having exam_score1 exam_score2 etc. since that could be confusing. I could imagine coming to a project a couple weeks or months later and not remembering if those fields were the same or actually represented different data.
3. Once the user deselects the second exam_score column from the Y-bar field, that calculated column is deleted. The field was purpose-specific and the user doesn't have multiple copies of columns hanging around.

Out of curiosity, what's the limitation in having perspective do something under the hood where it allows the user to select exam_score again from the dropdown and internally, it creates an alias named however perspective's internals require but keeps the display field to the user as exam_score?

[texodus]

Holding the Shift key while clicking the reset button will delete all expression columns (this isn't an complete answer, it's just a tip). Expression columns actually have no performance/memory cost if they are not "in" the query, when they are deselected they are not calculated. The fact that in the UI they need an extra "step" to be deleted is just a convenience so you don't have to re-type the expression.

Perspective internally uses column name string itself as the unique column identifier, so we cannot physically have duplicate column names, even as expression aliases. The impact of this design decision trickles all the way down to e.g. the persistence and internal wire formats, and would be a major breaking change.

Additionally, this is a feature breaking change in that many Perspective-based systems are designed around the contract of a viewer config JSON that works when the row data changes. As most data formats (even Arrow) are not structured with aliases in mind, Perspective cannot deterministically rehydrate a Viewer config for a dataset if there are duplicate column names because it can't tell which columns go with which config aliases.

Like I said, we have some UX features in the works to make this a little quicker, and it is (I think) possible to make this design mostly transparent - we can duplicate the column implicitly when you drag/drop to a "duplicate" slot e.g., but this requires some careful consideration of the drag/drop rules.

[mhkeller]

Makes sense. I think something where it kept a unique name internally but the same name display name for the user would be a good solution from a user perspective. My mentioning of cleaning up was purely from a user perspective and not related to performance.

[timkpaine]

duplicate names are also useful when you want to have e.g. a data column and a spark bar of the same column