Binned x/y encoding ignores scale domain parameter

[jheer]

When using a binned spatial encoding, custom scale domain settings seem to be ignored. The generated Vega pulls the scale domain from the data, not the provided custom domain.

Here is an example spec:

{
  "$schema": "https://vega.github.io/schema/vega-lite/v2.json",
  "data": {"values": [0.1, 0.6, 2.3]},
  "mark": "bar",
  "encoding": {
    "x": {
      "field": "data",
      "type": "quantitative",
      "bin": {"extent": [0, 10], "step": 0.5},
      "scale": {"domain": [0, 10]}
    },
    "y": {
      "aggregate": "count",
      "type": "quantitative",
      "scale": {"domain": [0, 2]}
    }
  }
}

I expect the x-axis domain to be [0, 10], but instead it is [0, 2.5].

[kanitw]

@yhoonkim I remember you work on this.

Can you remember why setting bin scale's domain is prevented?

https://github.com/vega/vega-lite/blob/master/src/compile/scale/domain.ts#L177

(It should be relatively easy to fix, but wanna understand if fixing doesn't cause other bad effects / regression.)

[yhoonkim]

@kanitw

I guess we prefer to control the scale by bin.extent instead of scale.domain.

But it's kinda weird that one of the bin parameter is outside of the bin object. (It's also more than just domain as it affects the bin transform that we perform before the scale.)
(#2791 (comment))

However,
it is still a problem when scale.domain and bin.extent are specified together.

My suggestion is ignoring scale.domain when the field is binned and throwing a warning message.
(#2429)

So we ignored scale.domain and force to use bin.extent with a warning message. + We decided to fix it later...

--

The above Jeff's spec generates x scale in vega spec like the below:

{
      "name": "x",
      "type": "linear",
      "domain": {
        "data": "data_0",
        "fields": [
          "bin_extent_0_10_step_0_5_data",
          "bin_extent_0_10_step_0_5_data_end"
        ]
      },
      "range": [
        0,
        {
          "signal": "width"
        }
      ],
      "zero": false
    }

However, bin_extent_0_10_step_0_5_data_end is 2.5 so it makes the x-domain [0,2.5]. (It is also related to a standard aggregation behavior [that] (vega/vega#983).) Thus, I think we have to keep using bin.extent as the domain but set vega's x-domain to be the user defined value ([0, 10]) instead of the value from the calculation ([0, 2.5).

[kanitw]

@yhoonkim can you help fix this then?

[yhoonkim]

@kanitw I'll try! To solve this, I need to hear your thoughts about the below specs and their outcomes. Could you tell me if they look OK?

1. Spec

{
    "$schema": "https://vega.github.io/schema/vega-lite/v2.json",
    "data": {"values": [0.1, 0.6, 2.3]},
    "mark": "bar",
    "encoding": {
      "x": {
        "field": "data",
        "type": "quantitative",
        "bin": {"extent": [0, 10], "step": 2},
        "scale": {"domain": [-1, 3]}
      },
      "y": {
        "aggregate": "count",
        "type": "quantitative",
        "scale": {"domain": [0, 2]}
      }
    }
  }

(In my opinion, the visualization looks bad, but the outcome of the spec is fine. I think Vega-Lite doesn't need to correct this bad, and users should be careful when specifying bin.extent and scale.domain together. )

2. Spec

{
    "$schema": "https://vega.github.io/schema/vega-lite/v2.json",
    "data": {"values": [0.1, 0.6, 2.3]},
    "mark": "bar",
    "encoding": {
      "x": {
        "field": "data",
        "type": "quantitative",
        "bin": {"extent": [0, 10], "step": 2},
        "scale": {"domain": [-10, 20]}
      },
      "y": {
        "aggregate": "count",
        "type": "quantitative",
        "scale": {"domain": [0, 2]}
      }
    }
  }

(In my opinion, It looks OK. If users want to draw ticks, they should specify bin.extent as the same as scale.domain. )

[kanitw]

1. is ok

2. is a bit weird that the axis values is only drawn from the bin value. In a way, if users customize scale.domain, then this is okay.

The problem is that axis's "values" is set to

"values": {
        "signal": "sequence(bin_extent_0_10_step_2_data_bins.start, bin_extent_0_10_step_2_data_bins.stop + bin_extent_0_10_step_2_data_bins.step, bin_extent_0_10_step_2_data_bins.step)"
      },

I think it's better if the start and stop are replaced with domain min and domain max.

but I'm not sure how to do that in Vega expression and maybe it's okay to just drop "values" in such case.

3. More importantly, this reminds me that users shouldn't have to set domain in addition to extent.

We should make scale domain depends on explicitly set bin's extent too. (This is even more important than correcting axis's "values".)

Basically,

{
    "$schema": "https://vega.github.io/schema/vega-lite/v2.json",
    "data": {"values": [0.1, 0.6, 2.3]},
    "mark": "bar",
    "encoding": {
      "x": {
        "field": "data",
        "type": "quantitative",
        "bin": {"extent": [-10, 10], "step": 2}
      },
      "y": {
        "aggregate": "count",
        "type": "quantitative",
        "scale": {"domain": [0, 2]}
      }
    }
  }

should have the domain = [-10, 10] nicely like this:

(I guess this isn't too magical and should be a reasonable default?)