search

ocean-eddy-cpt / gcm-filters-paper

Manuscript on spatial filtering method
1 stars 0 forks source link

tweak numerical instability figure #30

Closed NoraLoose closed 3 years ago

iangrooms commented 3 years ago

The three panels have massively different ranges: 10^21 for the first panel, 20,000 for the second panel, and 22 for the third panel. Can we get three different colorbars? I tried that unsuccessfully, so I don't know how hard it is to achieve with matplotlib.

Also, I just pushed a commit to fix the range of k on the y axes. It's now -127 to 128 and I just had to add ,extent=[0,127,128,-127].

NoraLoose commented 3 years ago

The three panels have massively different ranges: 10^21 for the first panel, 20,000 for the second panel, and 22 for the third panel. Can we get three different colorbars? I tried that unsuccessfully, so I don't know how hard it is to achieve with matplotlib.

Not hard at all - I just thought having one colorbar would enable a quicker visual comparison. But yes, it's not optimal to cut 17 orders of magnitude for the first panel (even though I would see 10^4, where the colorbar is saturated, as a blow-up already).

I can easily change this, so we have 3 different colorbars. Would you like to have the same colormap for all three, but with max value 10^21, 10^4, and 100 respectively?

iangrooms commented 3 years ago

Maybe 10^21, 10^4 and 20. I'm not sure what will look best. I trust you

NoraLoose commented 3 years ago

It is actually not that straightforward to show what is going on in this plot, when you use just linear or logarithmic colormaps (see my various attempts in the Roundoff.ipynb notebook that I just pushed). I had to break my vows and do a little colormap engineering. This is the best I came up with:

instability

iangrooms commented 3 years ago

Looks great! I'll pull this in and put it in the paper.