Skip to content

GaussianKDE

GaussianKDE

Bases: DensityCurve

Class for creating Gaussian Kernel Density Estimation (KDE) plotters. Use this class for univariate plots.

Kernel density estimation is a way to estimate the probability density function (PDF) of a random variable in a non-parametric way. The used gaussian_kde function of scipy.stats works for both uni-variate and multi-variate data. It includes automatic bandwidth determination. The estimation works best for a unimodal distribution; bimodal or multi-modal distributions tend to be oversmoothed.

PARAMETER DESCRIPTION
source

Pandas long format DataFrame containing the data source for the plot.

TYPE: pandas DataFrame

target

Column name of the target variable for the plot.

TYPE: str

feature

Column name of the feature variable for the plot, by default ''. Please note that this class is designed for univariate plots, so by default, all curves will be drawn on the same feature axis. If you want to draw curves on different feature axes, change the "ignore_feature" parameter to False and provide a feature column in the source DataFrame.

TYPE: str DEFAULT: ''

stretch

Factor by which the curve was stretched in height, by default 1.

TYPE: float DEFAULT: 1

height

Height of the KDE curve at its maximum, by default None.

TYPE: float | None DEFAULT: None

skip_na

Flag indicating whether to skip missing values in the feature grouped data, by default None - None, no missing values are skipped - 'all', grouped data is skipped if all values are missing - 'any', grouped data is skipped if any value is missing

TYPE: Literal['none', 'all', 'any'] DEFAULT: None

ignore_feature

Flag indicating whether the feature axis should be ignored. If True, all curves have base 0 on the feature axis, by default True

TYPE: bool DEFAULT: True

margin

Margin for the sequence as factor of data range (max - min ). If margin is 0, The two ends of the estimated density curve then show the minimum and maximum value. Default is 0.

TYPE: float DEFAULT: 0

fill

Flag whether to fill in the curves, by default True

TYPE: bool DEFAULT: True

agreements

Specifies the tolerated process variation for calculating quantiles. These quantiles are used to represent the filled area with different opacity, thus highlighting the quantiles.If you want the filled area to be uniform without highlighting the quantiles, provide an empty tuple. This argument is only taken into account if fill is set to True. The agreements can be either integers or floats, determining the process variation tolerance in the following ways: - If integers, the quantiles are determined using the normal distribution (agreement * σ), e.g., agreement = 6 covers ~99.75% of the data. - If floats, values must be between 0 and 1, interpreted as acceptable proportions for the quantiles, e.g., 0.9973 corresponds to ~6σ. - If empty tuple, the filled area is uniform without highlighting the quantiles.

Default is DEFAULT.AGREEMENTS = (2, 4, 6), corresponding to (±1σ, ±2σ, ±3σ).

TYPE: tuple[float, ...] or tuple[int, ...] DEFAULT: AGREEMENTS

cumulative

If True, the cumulative distribution (CDF) is plotted instead of the density (PDF), by default False.

TYPE: bool DEFAULT: False

target_on_y

Flag indicating whether the target variable is plotted on the y-axis, by default True.

TYPE: bool DEFAULT: True

color

Color to be used to draw the artists. If None, the first color is taken from the color cycle, by default None.

TYPE: str | None DEFAULT: None

n_points

Number of points the kernel density estimation and sequence should have, by default KD_SEQUENCE_LEN (defined in constants.py).

TYPE: int DEFAULT: KD_SEQUENCE_LEN

ax

The axes object for the plot. If None, the current axes is fetched using plt.gca(). If no axes are available, a new one is created. Defaults to None.

TYPE: Axes | None DEFAULT: None

visible_spines

Specifies which spines are visible, the others are hidden. If 'none', no spines are visible. If None, the spines are drawn according to the stylesheet. Defaults to None.

TYPE: Literal['target', 'feature', 'none'] | None DEFAULT: None

hide_axis

Specifies which axes should be hidden. If None, both axes are displayed. Defaults to None.

TYPE: Literal['target', 'feature', 'both'] | None DEFAULT: None

**kwds

Additional keyword arguments that have no effect and are only used to catch further arguments that have no use here (occurs when this class is used within chart objects).

DEFAULT: {}

Examples:

Apply to an existing Axes object:

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from daspi import GaussianKDE

fig, ax = plt.subplots()
colors = ('#1f77b4', '#ff7f0e', '#2ca02c')
df = pd.DataFrame(dict(
    x = ['first'] * 50 + ['second'] * 50 + ['third'] * 50,
    y = (
        list(np.random.normal(loc=3, scale=1, size=50))
        + list(np.random.normal(loc=4, scale=1, size=50))
        + list(np.random.normal(loc=2, scale=1, size=50)))))
for color, (name, group) in zip(colors, df.groupby('x')):
    kde = GaussianKDE(
        source=group, target='y', strategy='norm', agreements=(2, 4, 6),
        target_on_y=False, color=color, margin=0.3, ax=ax)
    kde(label=name)
handles, labels = ax.get_legend_handles_labels()
agreements = kde.agreements[::-1] * df.x.nunique()
labels = [f'{l} {a}σ' for l, a in zip(labels, agreements)]
ax.legend(handles, labels)

Apply using the plot method of a DaSPi Chart object:

import numpy as np
import daspi as dsp
import pandas as pd

df = pd.DataFrame(dict(
    x = ['first'] * 50 + ['second'] * 50 + ['third'] * 50,
    y = (
        list(np.random.normal(loc=3, scale=1, size=50))
        + list(np.random.normal(loc=4, scale=1, size=50))
        + list(np.random.normal(loc=2, scale=1, size=50)))))
chart = dsp.SingleChart(
        source=df,
        target='y',
        hue='x',
        target_on_y=False,
    ).plot(
        dsp.GaussianKDE,
        strategy='norm',
        agreements=(2, 4, 6), # std for area color opacities
        visible_spines='target',
        hide_axis='feature',
    ).label() # neded to label feature ticks

estimate

Estimate the kernel density of the target data, with the curve width controlled by margin.