(WIP) Filter and GroupBy Transforms

[timelyportfolio]

This pull makes the filter transform an actual part of Plotly and builds on the very fine transform work by @etpinard. The first filters available will be =, >, <, in, notin, within, notwithin (name changes welcome).

• add groupby to this pull

[etpinard]

@chriddyp would you mind testing this out in the plot.ly workspace?

[chriddyp]

does this work for datetime strings too?

[timelyportfolio]

will try to get an example up, but yes it does work with datetime strings. It does not seem to work though with Date and datetime strings.

[chriddyp]

Nice! This is started to look good. Here's a prototype of this filter transform in Plotly 2:

Here a few more things that I'll need to be able to integrate this into plotly 2:

• Date, number, category support with the same auto-type checking that the axes.type: '-' uses
• Support filtersrc to be an array instead of just trace.x or trace.y. In Plotly 2, this will usually be a column of data that isn't visualized in the plot itself
• Filtering out nested data arrays. like marker.color or marker.line.color. (Note that this is different than just filtering out all arrays since we don't want to fill out style keys like colorscale)

[etpinard]

@chriddyp

Date, number, category support with the same auto-type checking that the axes.type: '-' uses

Sounds good. I'm thinking the filter functions should respect the (maybe autotyped) axis types for data arrays that have a corresponding axis (e.g. x, y and z in 3D) and otherwise cast everything to numbers.

I'm thinking that the < and > filter for type: 'categories' axes should be determined by the categories order as used internally to map categories to axis values. For example,

var data = [{
  x: ['A', 'B', 'D'],
  transforms: [{
    type: 'filter',
    operation: '>',
    value: 'B'
  }]
}]

Plotly.plot(gd, data);

gd._fullData[0].x   // => ['D']

Support filtersrc to be an array instead of just trace.x or trace.y. In Plotly 2, this will usually be a column of data that isn't visualized in the plot itself

That sounds a bit odd. Why would anyone filter a trace with no x / y data? Do you want to use the filter transform for grids? In that case, you should use the filter module transform method with mock "data" without calling Plotly.plot.

Filtering out nested data arrays. like marker.color or marker.line.color. (Note that this is different than just filtering out all arrays since we don't want to fill out style keys like colorscale)

This one will be easily implemented after @monfera 's #936

[etpinard]

made obsolete by #936