boxfill missing value does not respect opacity

[doutriaux1]

import vcs
import cdms2

f=cdms2.open(vcs.sample_data+"/clt.nc")

s=f("clt")
x=vcs.init()
x.plot(s)
s2=cdms2.MV2.masked_greater(s,50.)
gm =x.createboxfill()
gm.missing=[50,50,50,50]
x.plot(s2,gm)

raw_input("press enteR")

gives:

[doutriaux1]

grey should be transparent

[durack1]

@doutriaux1 happy to test this out in my example case, so for isofill:

And for boxfill:

I'd also be curious about the isofill clipping along the Northern pole, this feature seems a weird artifact of the graphics method to me

[durack1]

@danlipsa it would be great to get your input on this, I was hoping to generate a correct figure (with the masking working correctly) by the end of this week

[sankhesh]

@doutriaux1 @durack1 Looking into this now

[durack1]

@sankhesh great, let me know when you have a branch to test..

[sankhesh]

@doutriaux1 @durack1

PR #123 fixes the issue

Here is clt with gm.missing = [50, 50, 50, 0]

and with gm.missing = [50, 50, 50, 50]

[doutriaux1]

awesome! thanks @sankhesh

[durack1]

@doutriaux1 great, once this is merged by @danlipsa will it be magically available through the nightly channel? I was hoping to get these figs done this week...

[durack1]

@danlipsa @doutriaux1 @sankhesh I'm also curious about the northern bound behaviour in the isofill example above. It seems that some boundary effect is impacting the isofill so that missing (or lower) values are found along the northern fringe, with boxfill this isn't an issue as there is no interpretation/interpolation of the data, it just plots cells as they are stored. I think the current isofill behaviour is wrong

[danlipsa]

@durack1 I don't really understand what is missing?

[danlipsa]

@durack1 Are these the black values along the northern bound?

[durack1]

@danlipsa exactly, so identical data yet you get the differences (the black values in isofill along the northern boundary) as per the screenshot below:

And this plot is isofill on top (with the cropping/interpolation issues) and boxfill on the bottom - identical data

[doutriaux1]

@danlipsa yes. It's odd, but I think it's the algorithm, not much we can do about it, high gradient around there

[danlipsa]

@durack1 Are the white values masked?

[durack1]

@doutriaux1 if you were to pad the boundary with the same value one pixel around the boundary (so obviously here it's 90N+), then you wouldn't get such an artifact.. And that is exactly what boxfill faithfully plots, isofill is attempting to pad with interpolated values (assuming there is no or missing value beyond the 90N bound) is my guess

[durack1]

@danlipsa nope, the masked values (in boxfill lower panel) are the grey colour

[danlipsa]

@durack1 Can we have the nc file and the script?

[durack1]

@danlipsa here you go, it's a single time slice.. I can provide the plotting script if you want that as well..
siconc_input4MIPs_SSTsAndSeaIce_CMIP_PCMDI-AMIP-1-1-1_gn_187001.nc.zip

[durack1]

@doutriaux1 it seems this change isn't available through nightly on conda the latest available build is 21st Jan 2017 (not 27/28th)

[durack1]

@sankhesh @danlipsa @doutriaux1 just to confirm, using the vcx-nox 2017.2.22-py27_0 uvcdat/label/nightly and vtk-cdat-nox 7.1.0.2017.1.20-py27_0 uvcdat/label/nightly yields the image below, which is exactly what I expected.. Thanks!

Compare the this boxfill image to the second image above