Smooth line in facet plot

I’m trying to generate a faceted scatter plot with a line for the mean. This code works:

scatter = alt.Chart(data).mark_point().encode(
    x="Position",
    y="Value",
    color="Cluster",
)

average = alt.Chart(data).mark_line(color="black").encode(
    x="Position",
    y="mean(Value)")

combined = scatter + average

combined.facet("Cluster")

However, as my data is sparse in some areas, I get very spiky lines that don’t capture the trend well. My quick fix is to round the values of data.Position for the line, which works okay, but I’m wondering if there’s a clever way to make use of statsmodel’s lowess smooth? I can only see doing that in pandas, adding a new column with smooths per cluster.

[jakevdp]

Sorry, lost track of this. You are computing the running average of all points; I think you want to compute the running average within each group, which you can do by adding a groupby argument to your window transform.

combined_facet = (scatter + average).facet(
    "Cluster:N"
).transform_window(
    running_avg="mean(Value)",
    frame=[-10,10],
    sort=[alt.SortField("Position","ascending")],
    groupby=['Cluster']
).configure_mark(opacity=0.8)

[kuchenrolle]

As I mentioned, I used the sort argument. For no sorting, however, can the average be higher than all data points.