frequencies questions

[jameshadfield]

cc @huddlej @trvrb @rneher

Some relevant questions (add more as needed):

• Should normalisation be the default?
• How should residual error be dealt with? Currently if the raw frequency for a pivot is <0.01 I collapse it to zero -- the alternative being to expand what is probably rounding / smoothing errors to what looks like 100% frequency. But this approach leads to sawtooth like displays:
  

[trvrb]

@jameshadfield and @rneher: I thought a bit here and I came up with the following proposal:

1. We properly normalize tip frequencies in augur before exporting to auspice. With all tips shown the frequencies display will look like:

2. We keep only "raw" frequencies and disable normalization. However, we dynamically rescale the y axis to fill the plot, so that zooming to a clade A1 would look like:

This should keep the app from getting into funny states and also keep the visualization well interpretable.

Even a very small clade ends up looking okay:

Edit: Also, just realized that this is the only natural way to do "frequencies" for something like Ebola. In this case, "raw" should have something proportional to cases on the y, rather than frequencies. This decision can be made on a per-pathogen basis in augur.

[rneher]

this looks good to me (even though I still don't fully understand the notion of a tip frequency)