parallel_coordinates is incorrect when using `color` and non-sorted index

[shane-kercheval]

This is somewhat of a weird bug, but I don't think it would be that uncommon of a use-case.

First let's see it working as expected.

If you run:

df = pd.DataFrame({
        'A': [1, 2, 3, 4],
        'B': [1, 2, 3, 4],
        'C': [1, 2, 3, 4],
    },
    index=[0, 1, 2, 3]
    #index=[3, 2, 1, 0]
)
df

px.parallel_coordinates(df)

you'll see

Then run:

px.parallel_coordinates(df, color='B')

and you'll see

Everything looks good so far.

But then change the index of the DataFrame from index=[0, 1, 2, 3] to index=[3, 2, 1, 0]

df = pd.DataFrame({
        'A': [1, 2, 3, 4],
        'B': [1, 2, 3, 4],
        'C': [1, 2, 3, 4],
    },
    #index=[0, 1, 2, 3]
    index=[3, 2, 1, 0]
)
df

px.parallel_coordinates(df)

This will return exactly the same graph as before. But when you use the color param, you get this

px.parallel_coordinates(df, color='B')

Which is incorrect. For example, instead of 1-1-1 (which is still a single row, despite the change in index), the line goes from 1-4-1.

[shane-kercheval]

So it seems like the code corresponding to the color variable is doing a sort or something that isn't retaining the correct order. A and C still have the correct associations, but B is reversed.

[Adrien-LUDWIG]

I have come into a similar case. I do not think it needs a new issue but this one may need a slight title update.

In my case, the index is sorted. When I plot the original dataframe, there is no problem. But when I plot a slice of the same dataset (keeping the index sorted), the values are disturbed. The problematic values are not necessarily the ones of the column used for coloring.

To reproduce

df = pd.DataFrame({
    'id': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
    'value': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
})

df2 = df[df["id"] % 2 == 0]

# With color
px.parallel_coordinates(df2, color="value")

#Without color
px.parallel_coordinates(df2)

[MarcoGorelli]

hi @shane-kercheval

is this your expected output?

[shane-kercheval]

Yes, thank you!

[MarcoGorelli]

Awesome! You should already be able to get the fix by using the 6.0 pre-release with the latest Narwhals version

pip install -U --pre plotly narwhals>=1.15.1

I'd be amazing if you could try it out and report any issues you encounter so they can be fixed before the final 6.0 release 🙏 ⭐