Data Binning question

[JobLeonard]

I need input on what is a sensible data representation.

I'm trying to make the legend accompanying the sparklines/scatterplots a bit "smarter", basically having sensible defaults (that you can override, don't worry):

(aside: I've used an interpolation tool to increase the solar9 gradient to a solar256 gradient, and also added the YlGnBu scale as a Heatmap2 option - being protanomalous the latter is easier to read for me)

The issue is that some metadata fields, like _LogCV or _tSNE1 & _tSNE2 tend to end up with (nearly) unique values for each column/row entry. This leads to these silly results:

(also, in the process of opening this issue I realised that gene data is no longer being sorted. Ignore that visual error for now, I'll fix that soon)

There's also the fact that zero values often representing missing data:

So I think that it would make sense to "bin" numerical data before creating a legend. Here are my thoughts so far:

• determine min/max values to determine dynamic range
  • ignore zero values for this purpose, since it's often used to represent "missing" data
• bin to N value ranges
  • determine if ranges needs to be log-scaled (I guess checking if the attribute name starts with _Log would work? And gene values?)
  • if zero-values are present they get their own bin
• display value ranges in legend (allow filtering on these ranges? This would somewhat complicate the code/data structure, but if you consider this worth it I'll build it)
  • color bins by mapping to the gradient palette

Is this a valid, clear way of representing the data, or would you prefer something else? How many bins would you consider sensible? Do you need filtering capabilities for these bins?

Thanks for your feedback,

/Job

[JobLeonard]

Conclusion after quick chat with Gioele: "Eh, fixing this would be nice to have, but not high priority"

• values like _LogCV are not that relevant compared to other data
• the heatmaps are still readable without a scale and this tool is more for exploration than rigorous data analysis anyway.