Model architecture: Add support for continuous color ramps

[kring]

E.g. http://ci.terria.io/master/#build/TerriaJS/test/init/csv.json and add CSV Testing -> Colors -> CSV with colorBins=0.

[nf-s]

#5426

[nf-s]

Related to #3524 SDMX 2021-02-23_Testing feedback:

Some heat maps don't have high contrast

• Not a major issue but some heat maps aren't displaying the contrast between low and high values on the map:
• E.g. Retail Trade in previous National Map: https://nationalmap.gov.au/#share=s-e1xho9kGoFg0hVO3LvFs0lF4TgW
• Same data in new National Map: http://ci.terria.io/sdmx-fixes-loading/#share=s-4lk6ZdzU1VcnSJIdTQIyCLP2CyU

This can be solved by using a continuous color ramp - i.e. not binning the data (#1585)

[nf-s]

From duplicate ticket: #1585

Rather than binning the area values, they could be mapped to a continuous interval. For example, we could:

• 'equalise' the value distribution through a non-linear mapping - eg compute (area rank)/(number of areas) to obtain a value between 0 and 1
• linearly map the values to the (0,1) interval by computing (y-y_min)/(y_max-y_min). This can be made more robust to data outliers by mapping between the nth and nth last percentiles, rather than the min and max.
• apply any other transforms such as truncated logarithm

We then apply the colour map function to each value to compute a RGB colour. There are many choices of such functions in the literature, but a perceptually linear colour map would be a principled choice - Matplotlib implements a bunch of (often piecewise-linear) functions that could be easily ported to javascript:
http://matplotlib.org/examples/color/colormaps_reference.html
https://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/_cm.py

[wibily]

considering clamping. E.g most data between 0-5 and a random 2000 value

[nf-s]

Closed by #5426