support cmap parameter for line plots

[knaaptime]

I have a dataframe data with three dimensions (x,y,z) and I'd like to create a line plot where the color of the line is set by a colormap rather than randomly-chosen colors.

currently I can do

data.hvplot(x='x', y='y', by='z', cmap='Reds')

which gives nearly the appropriate plot, except that the cmap parameter is ignored. This causes problems when z is a continuous variable rather than a categorical variable because in that case, a sequential colormap is necessary to shade each line appropriately (instead of the default categorical colormap). I'd like for the cmap parameter to be recognized in line graphs as it is with other types, e.g. scatter

for example using hvplot I can do

data.hvplot(x='distance', y='Index', by='year', cmap="reds")

which yields the top figure below (ignoring 'cmap`). Instead, I'd like to generate something like the bottom figure (created using Altair)

[maximlt]

This seems to me like a very nice feature to have in hvPlot! This is actually currently possible, using .opts and some HoloViews-fu, applying the options to the elements contained by the resulting NdOverlay:

import holoviews as hv
import hvplot.pandas
import numpy as np
import pandas as pd

df = pd.DataFrame({
  'x': list(np.arange(10)) * 6,
  'y': np.concatenate([np.arange(10) + x for x in range(6)]),
  'z': [l for l in range(0, 60, 10) for _ in range(10)],
})

df.hvplot.line('x', 'y', by='z').opts(hv.opts.Curve(color=hv.Palette('Reds')))

Any thought on this @jlstevens @jbednar ? This is right now in the Wishlist milestone so bound to best ignored indefinitely. I feel this should be supported by hvPlot but one could make the case that it's just a matter of better documenting .opts to hvPlot users.

By the way the fact that passing cmap is swallowed without any effect or error/warning is a bug, I'm going to open an issue for that.

[LecrisUT]

Does anyone have a quick update on the workaround? With the latest version it complains that hvplot.opts is not defined. It's holloviews not hvplot, my bad

[maximlt]

I hit this issue again when working on documenting hvPlot's compatibility with Pandas API.

import pandas as pd
import numpy as np

ts = pd.Series(np.random.randn(1000), index=pd.date_range('1/1/2000', periods=1000))
ts = ts.cumsum()
df = pd.DataFrame(np.random.randn(1000, 10), index=ts.index)
df = df.cumsum()
df.plot(colormap="cubehelix", backend='matplotlib')

The same code using cmap/colormap with hvPlot simply ignores the passed colormap:

This can be worked around by setting color with an hv.Palette object:

import hvplot.pandas
import holoviews as hv
import matplotlib
import pandas as pd
import numpy as np

# cubehelix palette only added when the matplotlib extension is loaded
hv.Palette.colormaps['cubehelix'] = matplotlib.colormaps['cubehelix']

ts = pd.Series(np.random.randn(1000), index=pd.date_range('1/1/2000', periods=1000))
ts = ts.cumsum()
df = pd.DataFrame(np.random.randn(1000, 10), index=ts.index)
df = df.cumsum()
df.hvplot(color=hv.Palette("cubehelix"))

---

The code I suggested a couple years ago can be simplified a bit:

import holoviews as hv
import hvplot.pandas
import numpy as np
import pandas as pd

df = pd.DataFrame({
  'x': list(np.arange(10)) * 6,
  'y': np.concatenate([np.arange(10) + x for x in range(6)]),
  'z': [l for l in range(0, 60, 10) for _ in range(10)],
})

df.hvplot.line('x', 'y', by='z', color=hv.Palette('Reds'))

[jbednar]

@maximlt , would you propose that cmap or colormap does the same thing as color here? If so that sounds good, and seems like a PR that would improve people's lives!

[maximlt]

Yes, and without having to use the hv.Palette object.

@Azaya89 is this something you recorded in your notebook that documents all the ways users can color things in hvPlot?

[Azaya89]

I think there are multiple issues here...

In the dummy dataframe below, the column being used to color the line plot is a numerical column ('z'):

import hvplot.pandas
import numpy as np
import pandas as pd

df = pd.DataFrame({
  'x': list(np.arange(10)) * 6,
  'y': np.concatenate([np.arange(10) + x for x in range(6)]),
  'z': [l for l in range(0, 60, 10) for _ in range(10)],
})
df.sample(5)

Initial problem -> cmap is ignored

df.hvplot.line('x', 'y', by='z', cmap='Reds')

However, if you set datashade=True in the plot, the cmap keyword is consumed (somewhat)

df.hvplot.line('x', 'y', by='z', cmap='Reds', datashade=True, dynspread=True, threshold=1)

df.hvplot.line('x', 'y', by='z', cmap='Set1', datashade=True, dynspread=True, threshold=1)

However, it doesn't make sense to use datashader for line plots just to get your preferred cmap used. So another way you can get hvplot to use the cmap is to plot with another element type - in this case, a points or scatter plot:

df.hvplot.points('x', 'y', color='z', cmap='Reds')

Notice however, that I used the color keyword instead of the by keyword, and also, you get a colorbar since you are coloring the plot using your numerical column. However, if I add a categorical column to the dataframe and use it to color the plot:

df['cat'] = 'cat' + df['z'].astype(str)
df.hvplot.points('x', 'y', color='cat', cmap='Reds')

Now you get a legend instead of a numerical colorbar. Doing the same with the original line plot will generate a ValueError:

df.hvplot.line('x', 'y', color='z', cmap='Reds')

ValueError [Call holoviews.ipython.show_traceback() for details]
Mapping a dimension to the "color" style option is not supported by the Curve element using the bokeh backend. To map the "z" dimension to the color use a groupby operation to overlay your data along the dimension.

In conclusion, I think the colormap you get depends on the type of plot you are constructing (line vs scatter/points), the data type of the dimension you are using to color the plot (numerical or categorical), and/or the keyword you are using (by vs color).