Altair: Missing values interpolation

[ADR777]

When I have a line (points=true) chart with years on the x-axis spanning from say 1975 to 2019 and have a value for every year, Altair joins all the points values by default, with a line. However, if there is a gap of several years, I would prefer that Altair not to fill in the gap between points. How is this best achieved?

[jakevdp]

I think what you're looking for is impute with a value of None. For example:

import pandas as pd
import altair as alt

data = pd.DataFrame({
    'x': [1, 2, 3, 5, 6, 7, 9, 10],
    'y': [1, 2, 3, 4, 5, 6, 7, 8],
})

alt.Chart(data).mark_line(point=True).encode(
  x='x',
  y=alt.Y('y',
    impute=alt.ImputeParams(
      value=None,
      keyvals=list(range(11))
    )
  )
)

More information about the impute transform in the docs: https://altair-viz.github.io/user_guide/transform.html#impute-transform

[ADR777]

Thank you Jake

The impute works until I select only the items that are imputed, Then it fails to work as seen in the pictures below. I can send my code and files?

[jakevdp]

Hard to tell without seeing your code, but I suspect you're using a filter transform to select lines. Try using a conditional opacity instead, so the dataset itself doesn't change.

[ADR777]

min = bore_df['Value'].min()
max = bore_df['Value'].max()

highlight = alt.selection(type='single', on='click',
                      fields=['Hole_Pipe'], nearest=True)

# This allows you to select in the legend
# multiple items with shift key held down    
selectionmulti = alt.selection_multi(fields=['Hole_Pipe'])

colors = alt.condition(selectionmulti,
                      alt.Color('Hole_Pipe:N', legend=None),
                      alt.value('lightgray'))

colores = alt.condition(selectionmulti,
                      alt.Color('Hole_Pipe:N', legend=None),
                      alt.value(None))

base = alt.Chart(bore_df).encode(
  x=('Date:T'), 
  #y='Value:Q',
  y=alt.Y('Value:Q',
          impute=alt.ImputeParams(value=None),
          scale=alt.Scale(domain=[min, max])),
  color=colores
  )


# opacity=alt.value(1) makes the circles not be opaque
# opacity=alt.value(0.1) makes the circles be opaque
# .transform_filter(selectionmulti) allows only the selected items
# in the legend to be displayed in the tooltips 
chartpoint = base.mark_point(point=True, size=50).encode(opacity=alt.value(1),
        tooltip=['GW_No', 'Hole_Pipe', 'Date', 'Value']).add_selection(
        highlight
  ).transform_filter(selectionmulti)  


chartline = base.mark_line().encode(
        size=alt.condition(~highlight, alt.value(1.2), alt.value(2.5))
        ).properties(title=bore, width=1100, height=570).interactive().transform_filter(selectionmulti)  


legends = alt.Chart(bore_df).mark_circle(size=300).encode(
        y=alt.Y('Hole_Pipe:N', impute=alt.ImputeParams(value=None), axis=alt.Axis(orient='right')),
        color=colors).add_selection(selectionmulti)


charty = ((chartline + chartpoint) | legends)

chart = (charty)


chart.save(results_dir + bore + '.html')

[ADR777]

This is what the data looks like

[jakevdp]

The impute transform only works if you tell it which values you want to impute. It can get these values two ways:

1. from other groups of data in the chart
2. from the keyvals argument to ImputeParams

If you're relying on (1) and you use a filter transform, then you change the values that are imputed depending on which values are present in the selected data. So you either need to (1) change the mechanism to use a conditional opacity rather than a filter transform as I mentioned above, or (2) explicitly set keyvals in your impute transform to specify which values you want imputed, as I showed in the code snippet above.

[jakevdp]

Regarding the documentation, "the Altair documentation is very hard to understand" is not helpful because it's not actionable. The best thing you could do is contribute to the documentation to make it easier to understand. The second best thing you could do is open issues pointing out specific problems.

[ADR777]

I do like Altair and feel its worth the investment in time to learn it. Its the best plotting library I have come across and I see a bright future for it.

[ADR777]

Thank You Jake for your reply. I have to spend time with my code and try to understand what you wrote and trial and error a way forward. Conditional opacity and how to use keyvals are not clear to me.

[jakevdp]

keyvals are the values at which the data is imputed. They can be specified two ways:

• explicitly, via the keyvals argument
• implicitly, via the presence of a value in the dataset visible to the chart element

When you use a filter transform, it changes the dataset visible to the chart element, changing the implicit keyvals, and thus changing imputed data.

So to fix your issue, you can either explicitly set keyvals, or avoid the filter transform.

By conditional opacity, I meant that you could replace this

chart.transform_filter(selectionmulti)

with this:

chart.encode(opacity=alt.condition(selectionmulti, alt.value(1), alt.value(0))

This lets you hide data without filtering it, so that the automatically determined keyvals will not change based on the selection.