centred text marks on sorted heat map

[Alcampopiano]

Hi Jake,

I assume that Altair 2.3 does not yet have the VL 3 sorting functionality mentioned here.

I am following a workaround involving making the x axis independent. This fixes the sorting but makes the text marks misaligned.

altair version 2.3

se=pd.DataFrame({'estimator': ['mean', 'mean', 'mean', 'mean',
                                'median', 'median', 'median', 'median',
                                'trim_mean', 'trim_mean', 'trim_mean', 'trim_mean'], 
                  'distribution': ['normal', 'lognormal', 'one-wild', 'slash',
                                  'normal', 'lognormal', 'one-wild', 'slash',
                                  'normal', 'lognormal', 'one-wild', 'slash'],
                  'se': [.1, .4658, 1.90, np.nan,
                        .138, .1727, .1679, .7048,
                        .113, .177, .1433, .9649]})


c=alt.Chart().encode(
    y=alt.Y('estimator'),
    tooltip=['estimator', 'distribution', 'se']
)

a=c.mark_rect().encode(
    x=alt.X('distribution', sort=['normal', 'lognormal', 'one-wild', 'slash'], axis=alt.Axis(labelAngle=0)),
    color=alt.Color('se', scale=alt.Scale(domain=[0,.5]))
)

b=c.mark_text(align='center').encode(
    x=alt.X('distribution', sort=['normal', 'lognormal', 'one-wild', 'slash'], 
            axis=alt.Axis(title=None, labels=False, ticks=False)),
    text='se',
)

alt.layer(a,b,data=se).resolve_scale(x='independent')

Is there a way to centre the text while maintaining the custom sort?

[jakevdp]

I assume that Altair 2.3 does not yet have the VL 3 sorting functionality mentioned here.

No, Altair 2.3 is still tied to Vega-Lite 2.6, as Vega-Lite 3 has yet to be released:

>>> import altair as alt
>>> alt.VEGALITE_VERSION
'2.6.0'

Regarding your question, I'm not sure how to answer it. When I run your code I get the following:

[jakevdp]

Here is an example that is similar to what you want: starting from your code snippet, I removed the resolve_scale(x='independent') and added baseline='middle' to the text mark:

import altair as alt
import numpy as np
import pandas as pd

se=pd.DataFrame({'estimator': ['mean', 'mean', 'mean', 'mean',
                                'median', 'median', 'median', 'median',
                                'trim_mean', 'trim_mean', 'trim_mean', 'trim_mean'], 
                  'distribution': ['normal', 'lognormal', 'one-wild', 'slash',
                                  'normal', 'lognormal', 'one-wild', 'slash',
                                  'normal', 'lognormal', 'one-wild', 'slash'],
                  'se': [.1, .4658, 1.90, np.nan,
                        .138, .1727, .1679, .7048,
                        .113, .177, .1433, .9649]})


c=alt.Chart().encode(
    y=alt.Y('estimator'),
    tooltip=['estimator', 'distribution', 'se']
)

a=c.mark_rect().encode(
    x=alt.X('distribution', sort=['normal', 'lognormal', 'one-wild', 'slash'], axis=alt.Axis(labelAngle=0)),
    color=alt.Color('se', scale=alt.Scale(domain=[0,.5]))
)

b=c.mark_text(align='center', baseline='middle').encode(
    x=alt.X('distribution', sort=['normal', 'lognormal', 'one-wild', 'slash'], 
            axis=alt.Axis(title=None, labels=False, ticks=False)),
    text='se',
)

alt.layer(a,b,data=se)

[Alcampopiano]

Thanks Jake. Yes, I should have mentioned that my above example was made in the traditional notebook (so I can use RISE). Perhaps that is why it shows up oddly. In JL I get the same as you.

Thanks again.

[jakevdp]

It might be that you're using an older version of the vega extension in the notebook, in which case an older version of vega-lite is being used to render your chart.

[jakevdp]

The rendering should be identical between notebook and jupyterlab if the vega extensions are up to date.

[Alcampopiano]

Hmm, I have Vega-Lite 2.6.0 and Altair 2.3.0 in the notebook. Does that tell us that I have the vega extension up-to-date?

[jakevdp]

The relevant package for the rendering of charts is not Altair, but the notebook vega extension.

It's actually quite difficult to figure out what vega-lite version is being used by the notebook to render your charts (see vega/ipyvega#97 for the open issue on this subject... cc/ @domoritz: this is an example of a situation where that enhancement would be helpful to users).

You need to update the vega package in order to use the most recent vega-lite version... version 1.4 is the most recent version.

[Alcampopiano]

Thanks Jake. Will do.