uadapy.plotting.shepard_diagram module

uadapy.plotting.shepard_diagram.custom_wasserstein(dist1, dist2)

uadapy.plotting.shepard_diagram.load_iris()

uadapy.plotting.shepard_diagram.plot_shepard_diagram(distributions_hi, distributions_lo, n_samples=1000, seed=55, metric='kl', labels=None, show_plot=False)

Plot a Shepard diagram to assess how well pairwise distances between distributions in a high-dimensional space are preserved in a reduced (low-dimensional) space. Supports predefined or custom distance metrics.

Parameters:

• distributions_hi (list of uadapy.Distribution or uadapy.Distribution) – The original (high-dimensional) distributions.

• distributions_lo (list of uadapy.Distribution or uadapy.Distribution) – The reduced (low-dimensional) distributions obtained via dimensionality reduction.

• n_samples (int, optional) – Number of samples to draw per distribution when using sample-based distance metrics. Default is 10000.

• seed (int, optional) – Seed for the random number generator to ensure reproducibility of sampling. Default is 55.

• metric (str or callable, optional) –

  Distance metric to use for computing pairwise distances between distributions. If a string, choose from: - “kl” : KL divergence between Gaussian approximations. - “wasserstein” : Average Wasserstein distance across dimensions. - “mean” : Euclidean distance between distribution means. If a callable, it must accept the following arguments:

  • dist1: uadapy.Distribution

  • dist2: uadapy.Distribution

  and return a scalar distance value. Default is “kl”.

• labels (list of str, optional) – List of class names for the distributions. If provided, labels like “class A vs class B” will be used to annotate points in the plot. If None, numeric indices will be used.

• show_plot (bool, optional) – If True, displays the plot. Default is False.

Returns:

• matplotlib.figure.Figure – The figure object containing the Shepard diagram.

• matplotlib.axes.Axes – The Axes object used for the plot.