How to create a histogram of non-numeric data used for a scale/legend?

[David]

I'm using Vega 3 and have data in the following format:

[{  "val": [ "X1",  "X2" ] },
 {  "val": [ "X2" ] },
 {  "val": [ "X1", "X2", "X3" ] } ]

I'd like to compute a histogram on val, take the top sorted histogram elements, and use the results for a legend and/or a scale range used in an axis.
In the above example, the result would be X2:3 and X1:2 with a top cutoff of 2.

How can this be done (without actually writing a new transform function)?

Thanks!

[Roy I]

The data for Vega v3 must be in a format that Vega v3 supports (see: https://vega.github.io/vega/docs/data/), e.g.

[ {"val": "X1"},

{"val": "X2"},

{"val": "X2"},

{"val": "X1"},

{"val": "X2"},

{"val": "X3"}

]

or:

[ "X1", "X2", "X2", "X1", "X2", "X3" ],

Here is a working Vega v3 spec for plotting ranked frequency bar chart with cutoff using your example data (works in new Vega v3 on-line editor: https://vega.github.io/new-editor/?mode=vega)

Vega spec (v3.0.0-beta.38)

---------------------------------------

{

  "$schema": "https://vega.github.io/schema/vega/v3.0.json",

  "width": 500,

  "height": 300,

  "padding": 5,

 

   "data": [

    { "name": "data_input",

      "values": [ {"val": "X1"},

{"val": "X2"},

{"val": "X2"},

{"val": "X1"},

{"val": "X2"},

{"val": "X3"}

]

    },

{ "name": "data_freq",

 "source": "data_input",

 "transform": [

  {

"type": "aggregate",

"groupby": ["val"],

"fields": ["val"],

"ops": ["count"],

"as": ["frequency"]

}

]

    },

{ "name": "data_freq_rank_filtered",

 "source": "data_freq",

 "transform": [

{ "type": "collect",

"sort": { "field": "frequency",

"op": "max",

"order": "descending"

}

},

{ "type": "rank",

"field": "val"

},

{ "type": "filter",

"expr": "datum.rank < 2"

}

]

    }

  ],

  

   "scales": [

   {  "name": "scale_x",

     "type": "band",

     "range": "width",

     "domain": { "data": "data_freq_rank_filtered",

       "field": "val"

},

"padding": 0.5

    },

    { "name": "scale_y",

      "type": "linear",

      "range": "height",

      "domain": { "data": "data_freq_rank_filtered",

       "field": "frequency"

       },

      "zero": true, 

   "nice": true

    }

  ],

  "axes": [

    { "orient": "bottom", "scale": "scale_x"},

    { "orient": "left", "scale": "scale_y", "tickCount": 3}

  ],

  "marks": [

        { "type": "rect",

          "from": {"data": "data_freq_rank_filtered"},

          "encode": {

            "update": {

           "fill": {"value": "steelblue"},

         "x": {"scale": "scale_x", "field": "val"},

         "width": {"scale": "scale_x", "band": 1},

         "y": {"scale": "scale_y", "field": "frequency"},

         "y2": {"scale": "scale_y", "value": 0}

            }

          }

        }

]

[Roy I]

By the way...

"Histograms are sometimes confused with bar charts. A histogram is used for continuous data, where the bins represent ranges of data, while a bar chart is a plot of categorical variables."

https://en.wikipedia.org/wiki/Histogram

On Tuesday, June 20, 2017 at 1:21:14 PM UTC-4, David wrote:

[David]

Thanks, Roy.
Problem is that the data format is as I specified, and there doesn't seem to be a standard transform to make the change - otherwise, yes, using aggregate would work.
Might have no other option than writing my own transform which changes the pulse.

[Roy I]

Here is an example javascript function that converts your data for Vega:

----------------------------

function convertData(input){

var result = [];

var arr;

for (var i = 0; i < input.length; i++){

arr = input[i]["val"];

for (var j = 0; j < arr.length; j++){

result.push({"val": arr[j]});

}

}

return result;

}

// test

var rawData = [{"val": ["X1","X2"]}, {"val":["X2"]}, {"val": ["X1","X2","X3"]}]; 

var myData = convertData(rawData);

console.log(JSON.stringify(myData));

// [{"val":"X1"},{"val":"X2"},{"val":"X2"},{"val":"X1"},{"val":"X2"},{"val":"X3"}]

On Tuesday, June 20, 2017 at 1:21:14 PM UTC-4, David wrote:

[David]

Thanks again :) The data for the histogram is reactive to events which cause filters on the upstream data, so I can't precalculate  the histogram or make the data conversion.
For example, a legend being clicked filters out some data, and the histogram must be computed on what is left.
I might be able to precalculate the conversion into another data collection, then do lookups inside the filter - though this may greatly impact performance.
Still trying to write a custom transform - thought this is coming to be a great challenge in and of itself :)