background_gradient vmin and vmax

[alimanfoo]

Working with the new style API, I would like to turn several columns of a dataframe into something like a heatmap. I would like to use the same range for all columns in the subset, i.e., the colour map should take a minimum and maximum value I specify, regardless of what actual range each column has. Currently using the "min" and "max" this is not intuitive. I would much prefer to be able to specify explicit vmin and vmax values, analogous to the seaborn.heatmap function.

Thanks for the style API, extremely useful.

[TomAugspurger]

That's what I wanted initially as well, but didn't have time to do it properly before the last release. I wonder if we should just require seaborn for those parameters, and use its heuristics for determining a good colormap and handling vmin/vmax.

[mpschr]

Hi,

I have some feedback which IIUC is related to what @alimanfoo is saying. The background_gradient as well as the bar style methods should have the same axis options as the highlight_max method:

axis: int, str, or None; default None

It can be very misleading if the color scale is only applied to each column separately - particulary when I specify a subset of columns (which represent the same measure) I expected them to be represented with the same color scale by default. I tried to set axis=None for the above mentioned methods, but unluckily an error is being raised from that.
I get that the feature is new and hopefully this will be possible soon. Apart from my negative criticism: Kudo's for the good work.

[interrogator]

Couldn't agree more with @mpschr --- axis = None should even be the default. Anybody know a workaround for now?

[davidbrochart]

@alimanfoo I'm not sure it applies directly to your problem, but to get the equivalent of vmin/vmax for the colormap, I do the following:

# given df, cmap, vmin, vmax:
df_data = df.copy()
df = df.clip(vmin, vmax)
df = df.style.background_gradient(cmap=cmap)
df.data = df_data

[soxofaan]

FYI: axis = None support is being added under PR #21259

[interrogator]

Thanks everyone, much appreciated.