RGB

[jsignell]

I was working with an xarray.DataArray with shape MxNx3 stored as a dask.array:

>>> im
<xarray.DataArray 'getitem-5f1ebf179271ce04f7cbd4e90ec207be' (y: 256, x: 256, band: 3)>
dask.array<shape=(256, 256, 3), dtype=uint8, chunksize=(256, 256, 3)>
Coordinates:
  * y        (y) int64 0 1 2 3 4 5 6 7 8 ... 247 248 249 250 251 252 253 254 255
  * x        (x) int64 0 1 2 3 4 5 6 7 8 ... 247 248 249 250 251 252 253 254 255
  * band     (band) int64 0 1 2

There are some issues that need to be resolved with holoviews first (holoviz/holoviews#3311), but then it would be nice to implement RGB in hvplot.

This would be particularly nice if the intake-xarray image plugin (intake/intake-xarray#25) gets in since we could then support RGB plotting in the intake plotting API.

For a comparison with matplotlib and full traceback:
https://gist.github.com/jsignell/4eeef0850708625e0b444db34600ea98

[scottyhq]

Thanks for adding this @philippjfr, it seems hvplot.rgb() doesn't work with DataArray types only DataSet:

import xarray as xr
import holoviews as hv
import hvplot

da = xr.open_rasterio('https://s3.amazonaws.com/earth-data/UCMerced_LandUse/Images/airplane/airplane00.tif')

# works:
da.plot.imshow()
hv.RGB(da.transpose('y','x','band').data).options(invert_yaxis=True) #works

# errors with DataArray
#hv.RGB(da.transpose('y','x','band').data, bounds=(0,256,0,256)).options(invert_yaxis=True) #error (ZeroDivisionError: float division by zero)
#da.hvplot.rgb() #error (KeyError: None)
#da.hvplot.rgb(x='x', y='y', bands='bands') #error (KeyError: None)

# rgb() works with Dataset instead of DataArray:
ds = da.to_dataset(name='z')
ds.hvplot.rgb(z='z', bands='bands') 

[philippjfr]

Thanks @scottyhq, I'll have a look and push a micro release out asap.

[philippjfr]

Fixed in #138.

[scottyhq]

Great! thanks for such a speedy fix. I also noticed a possible bug trying to browse multiple thumbnails with different sizes. Seems that the 'band' coordinate has to be first in order for the following to work. Might want to add another unit test for this case:

# Gives Error: WARNING:root:dynamic_operation: Callable raised "IndexError('index 3 is out of bounds for axis 0 with size 3')".
url1 = 'https://s3-us-west-2.amazonaws.com/landsat-pds/c1/L8/047/027/LC08_L1TP_047027_20130421_20170310_01_T1/LC08_L1TP_047027_20130421_20170310_01_T1_thumb_large.jpg'
da1 = xr.open_rasterio(url1)
url2 = 'https://s3-us-west-2.amazonaws.com/landsat-pds/c1/L8/047/027/LC08_L1TP_047027_20130523_20170310_01_T1/LC08_L1TP_047027_20130523_20170310_01_T1_thumb_large.jpg'
da2 = xr.open_rasterio(url2)
ds = xr.concat([da1,da2], 'time')
ds = ds.assign_coords(time=[0,1]).to_dataset(name='z')
print(ds)
ds.hvplot.rgb('x','y', z='z',bands='band',groupby='time', width=700, height=500)

# Works if 'band' is first coordinate
coords = dict(band=(['band'], ds.coords['band'].values),
              time=(['time'], ds.coords['time'].values),
              y=(['y'], ds.coords['y'].values),
              x=(['x'], ds.coords['x'].values),
              )

ds = xr.Dataset({'z': (['time','band','y','x'], ds['z'].data.astype('uint8'))}, coords=coords)
print(ds)
ds.hvplot.rgb('x','y', z='z',bands='band',groupby='time', width=700, height=500)

[philippjfr]

I think the issues have now been resolved, please reopen if issues persist.