Widget that switches hvplot.points color value between categorical and numeric type columns yields AttributeError

Panel
, ,
1

Posting this here because I’m not sure where the bug lies. I suspect Panel’s interaction with Bokeh is the problem.

import holoviews as hv
import pandas as pd
import numpy as np
import panel as pn
import hvplot.pandas
from holoviews.streams import Selection1D

data = {'X' : np.random.randn(100),
        'Y' : np.random.randn(100),
        'colorNum' : np.random.randint(0,10,100),
        'colorNum2' : np.random.randint(0,10,100),
        'colorCat' : list('ABCDEFGHIJ')*10}

df = pd.DataFrame(data)
df

image

Plotting this and switching between the two numerical columns works fine:

color_by_select = pn.widgets.Select(name='color by', options=['colorNum', 'colorNum2', 'colorCat'], value='colorNum')

def points(ds, colorby):
    return ds.data.hvplot.points(x='X', y='Y', color=colorby)

pn.Column(color_by_select,
          hv.Dataset(df).apply(points, colorby=color_by_select))

image
image901×492 32.4 KB

But a long error message is produced if I try to switch to coloring by colorCat:

Traceback (most recent call last):
  File "/home/popos/miniconda3/envs/maple/lib/python3.8/site-packages/holoviews/plotting/bokeh/element.py", line 1446, in _update_glyphs
    style = self._apply_transforms(element, data, ranges, style)
  File "/home/popos/miniconda3/envs/maple/lib/python3.8/site-packages/holoviews/plotting/bokeh/element.py", line 1136, in _apply_transforms
    cmapper = self._get_colormapper(v, element, ranges,
  File "/home/popos/miniconda3/envs/maple/lib/python3.8/site-packages/holoviews/plotting/bokeh/element.py", line 1922, in _get_colormapper
    opts = {k: opt for k, opt in opts.items()
  File "/home/popos/miniconda3/envs/maple/lib/python3.8/site-packages/holoviews/plotting/bokeh/element.py", line 1923, in <dictcomp>
    if getattr(cmapper, k) != opt}
  File "/home/popos/miniconda3/envs/maple/lib/python3.8/site-packages/bokeh/core/has_props.py", line 264, in __getattr__
    self._raise_attribute_error_with_matches(name, properties)
  File "/home/popos/miniconda3/envs/maple/lib/python3.8/site-packages/bokeh/core/has_props.py", line 272, in _raise_attribute_error_with_matches
    raise AttributeError(f"unexpected attribute {name!r} to {self.__class__.__name__}, {text} attributes are {nice_join(matches)}")
AttributeError: unexpected attribute 'factors' to LinearColorMapper, possible attributes are domain, high, high_color, js_event_callbacks, js_property_callbacks, low, low_color, name, nan_color, palette, subscribed_events, syncable or tags

Any way around this other than manually adding colors to a new column? If I have to do that, what would be the fastest way? Like could I just do essentially what Bokeh is doing in the backend but manually somehow?

Happy to repost this somewhere else if that would be better

Thanks

2

Hi @gordonrix

After some investigation I believe its a bug and have reported it here DynamicMap Issue: AttributeError: unexpected attribute ‘factors’ to LinearColorMapper

You can change to this implementation which works.

import holoviews as hv
import pandas as pd
import numpy as np
import panel as pn
import hvplot.pandas
from holoviews.streams import Selection1D

data = {'X' : np.random.randn(100),
        'Y' : np.random.randn(100),
        'colorNum' : np.random.randint(0,10,100),
        'colorNum2' : np.random.randint(0,10,100),
        'colorCat' : list('ABCDEFGHIJ')*10}

df = pd.DataFrame(data)

color_by_select = pn.widgets.Select(name='color by', options=['colorNum', 'colorNum2', 'colorCat'], value='colorNum')

def points(data, colorby):
    return data.hvplot.points(x='X', y='Y', c=colorby)

plot = pn.bind(points, df, color_by_select)

pn.Column(color_by_select, plot).servable()

image
image1282×493 32.3 KB

1 Like
3

Amazing, thank you so much!

1 Like
4

@Marc this workaround may not fully work for me actually. I’m having a bit of trouble knowing how to work with pn.bind. I need to be able to access the selection from the bound plot as a Selection1D object. Is that possible? My crude attempt produced an error:

import holoviews as hv
import pandas as pd
import numpy as np
import panel as pn
import hvplot.pandas
from holoviews.streams import Selection1D

data = {'X' : np.random.randn(100),
        'Y' : np.random.randn(100),
        'colorNum' : np.random.randint(0,10,100),
        'colorNum2' : np.random.randint(0,10,100),
        'colorCat' : list('ABCDEFGHIJ')*10}

df = pd.DataFrame(data)

color_by_select = pn.widgets.Select(name='color by', options=['colorNum', 'colorNum2', 'colorCat'], value='colorNum')

def points(data, colorby):
    return data.hvplot.points(x='X', y='Y', c=colorby)

plot = pn.bind(points, df, color_by_select)

index_stream = Selection1D(source=plot)

pn.Column(color_by_select, plot)

AttributeError: 'function' object has no attribute '_plot_id'

Oddly, this error persists even after commenting out the offending line (index_stream) until I restart my kernel.

5

Managed to get around this by making a basic static plot and using that as a selection stream source prior to passing it to the bind function, which combines it with a dynamic plot. I think I may take a performance hit though for rerunning the whole pipeline and rerendering every time I change one of the many widgets that I’m using, as opposed to what I think .apply does which is to only modify the already existing plot. Hopefully that bug gets patched soon.

1 Like