Help requested in contributing ECDF statistics to the Seaborn Objects

[JoostSP]

Hi,

I tried making a Seaborn Object statistic, that will allow people to create Emperical Cumulative Density Function plots.

My code is:

import seaborn.objects as so
import dataclasses
import typing
import numbers
import pandas as pd
import numpy as np

@dataclasses.dataclass
class ECDF(so.Stat):
    """Empirical cumulative distribution function, global or group-aware."""
    
    group_by_orient: typing.ClassVar[bool] = False

    @staticmethod
    def _compute_ecdf(data: pd.DataFrame, orient: str, orient_range: typing.Tuple[typing.Any, typing.Any]):
        var = {"x": "y", "y": "x"}[orient]
        orient_group_range = data[orient].min(), data[orient].max()
        return (
            data
                .dropna(subset=[orient])
                .sort_values(orient)
                .reset_index(drop=True)
                .pipe(lambda df: df.assign(**{
                    var: pd.Series(range(1, len(df) + 1), dtype=float) / len(df)
                }))
                .pipe(                        # Add start and end to the ecdf curves, so they plot nicely when more than one group is plotted
                    lambda df:
                        pd.concat(
                            [df.head(1).assign(**{orient: orient_group_range[0], var: 0.0}), df],
                            ignore_index=True
                        )
                )
                .pipe(                        # Add start and end to the ecdf curves, so they plot nicely when more than one group is plotted
                    lambda df:
                        pd.concat(
                            [df.head(1).assign(**{orient: orient_range[0], var: 0.0}), df] if orient_range[0] < orient_group_range[0] else
                            [df],
                            ignore_index=True
                        )
                )
                .pipe(                        # Add start and end to the ecdf curves, so they plot nicely when more than one group is plotted
                    lambda df:
                        pd.concat(
                            [df,  df.tail(1).assign(**{orient: orient_range[1], var: 1.0})]  if orient_group_range[1] < orient_range[1] else
                            [df],
                            ignore_index=True
                        )
                )
        )

    def __call__( self, data: pd.DataFrame, groupby, orient: str, scales: typing.Dict[str, object] = None ) -> pd.DataFrame:
        orient_range = data[orient].min(), data[orient].max()
        grouping_vars = [str(v) for v in data if v in groupby.order]
        
        if len(grouping_vars) == 0:
            return self._compute_ecdf(data, orient, orient_range)
        else:
            return (
                data
                    .groupby(by=grouping_vars, observed=True, group_keys=True)[data.columns]
                    .apply(lambda df: self._compute_ecdf(df, orient, orient_range))
                    .reset_index(drop=True)
            )

And could be used with:

import seaborn as sns

(
    so.Plot(diamonds_df, x='price')
        .add(so.Line(), ECDF())
        .limit(x=(0, max(diamonds_df.price)+1))
        .label(x='Price', y='Cumulative density',
               title='Empirical Cumulative Distribution Function (ECDF) plot of Price Distribution')
        .show()
)

I am however not very proficient in the whole code contribution/github pull request/CI-pipelines/automated test process.

Would somebody be willing to help me with the process of contributing this statistic to Seaborn Objects?

Thanks!
Joost

[JoostSP]

There is overlap with the code of issue 3250.