Set bin size to 1 for binX

[mkfreeman]

Is it possible to specify to Plot.binX() that I want only one observation per bin? For context, I want to create a bar chart of the value column in my dataset for each date. A few tricky pieces:

• Some dates are missing, so I can't simply use barY
• I don't want Plot to do any aggregation of the values, I want to see the exact values in the dataset (so hinting at a number of intervals isn't sufficient)
• The time interval in the data may change different (e.g., it may be each day, hour, etc.) so I can't simply specify the interval option

It seems that a rect mark with binning is the appropriate method, but I would want to ensure that there was only one observation per bin.

See a few approaches in this notebook: https://observablehq.com/d/4a1a0770b9e533b1

[Fil]

Fun. There is no solution if two data points have the same exact date. So you have to make the assumption that they are all different.

Then I suppose that the desired interval can be defined as a heuristic, such as the minimum of the differences between consecutive dates, maybe rounded to account for durations that overlap the DST:

interval = {
  const m = d3.min(
    d3.pairs(d3.sort(Plot.valueof(fakeData, "date")),
    ([a, b]) => Math.round((b - a) / 60000) // minutes
  );

  return m < 2
    ? d3.utcMinute
    : m < 10
    ? d3.utcMinute.every(m)
    : m < 60
    ? d3.utcMinute.every(m)
    : m < 23 * 60
    ? d3.utcHour.every(Math.round(m / 60))
    : m < 23 * 60 * 7
    ? d3.utcDay
    : m < 24 * 60 * 27
    ? d3.utcWeek
    : m < 24 * 60 * 364
    ? d3.utcMonth
    : d3.utcYear;
}

I haven't really tested and it's incomplete (milliseconds, quarters, centuries etc)

[mkfreeman]

Thanks @Fil 🫶 !