Blue plot

[hoxbro]

Noticed a blue plot on the holoviews documentation page here. It could be related to holoviz/datashader#1051.

[jbednar]

Yes, the blue is fairly distracting here, and I think we should do the same thing we do in Datashader, which is to make a distinction between no data (zero counts show up transparent) and data (non-zero counts are mapped to the colorbar). Bokeh has options for this, but I can't seem to make them work:

import pandas as pd, numpy as np, holoviews as hv
hv.extension('bokeh')

taxi_dropoffs = {hour:arr for hour, arr in np.load('../assets/hourly_taxi_data.npz').items()}
bounds = (-74.05, 40.70, -73.90, 40.80)

image = hv.Image(taxi_dropoffs['0'], ['lon','lat'], bounds=bounds)
image.opts(width=450, tools=['hover'], clipping_colors={'NaN': 'transparent'}, colorbar=True)
image

image.opts(nodata=0)

image.data = np.where(image.data == 0, np.nan, image.data)
image

We'd want it to look like the last image here, but to get that to work I had to manipulate the NumPy array directly; shouldn't it work to just specify nodata?

[hoxbro]

This works without manipulating the array directly image.redim.nodata(z=0)

[jbednar]

Thanks! @jlstevens , any idea how nodata=x is supposed to be used if not like this?

[jlstevens]

image.opts(width=450, tools=['hover'], colorbar=True, nodata=np.nan) should work as nodata is both an option on dimensions and a plot option for images. I am assuming the image has nans that need to be transparent and not zeros: just set nodata to whatever value needs to be transparent so you don't need to set clipping_colors explicitly.

[jbednar]

The image has zeros, not nans, and nodata=np.nan fails in any case:

[jbednar]

The question is how to make the zeros in the image map to NaNs; nodata=0 isn't meant to do that? The docs seem to claim that it is:

nodata:            Optional missing-data value for integer data.
                   If non-None, data with this value will be replaced with NaN so
                   that it is transparent (by default) when plotted.

[jbednar]

Ah, that's it: the data is float32, with "0." instead of "0". I would have thought nodata would apply to floats too, but it does work if I convert to int32:

taxi_dropoffs = {hour:arr.astype('int32') for hour, arr in np.load('../assets/hourly_taxi_data.npz').items()}
...
image.opts(nodata=0)

It's slightly annoying that the hover shows NaN instead of 0, but I guess Bokeh is just filling with NaNs internally.

[jlstevens]

Ok, I am remembering this now...

I think we decided to focus on counts as those are integers (the main use case) and comparing floats is always tricky. I assume the floats here are due to the way the data is stored, because otherwise I would expect integers (unless this is using a mean aggregator for instance).

[jbednar]

Right. I don't know where this data came from, but it does seem like it should have been int. Either need to fix the data, or add one of the above fixes to the notebook. Hmm.

[jbednar]

This is fixed on http://dev.holoviews.org/getting_started/Introduction.html and should appear on the main site when the docs are rebuilt on release.

[github-actions]

This issue has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.