UADAPy - Uncertainty-aware Data Analysis with Python

UADAPy is a Python package to support an easy analysis of uncertain multivariate data.

The library provides:

• a unified Distribution class that wraps around various other distribution types such as scipy.stats.<class>
• implementations of uncertainty-propagating visualization algorithms, e.g., UAPCA, UAMDS, UASTL
• simple plotting API that already contains all the boilerplate code to visualize distributions
  • specializes in iso-contour plotting for distributions (determines probability densities that correspond to specific quantiles, automatic grid positioning, orientation, and sizing for density sampling)
• uncertain/distributional datasets, e.g. Student Grades dataset

Installation

The package is available through PyPI and can be installed via pip

pip install uadapy

To get bleeding edge features, you can also install it from git

pip install git+https://github.com/UniStuttgart-VISUS/uadapy@<ref>

Documentation

You can find the documentation here: https://unistuttgart-visus.github.io/uadapy/ It also contains an overview of all supported methods.

Usage Example

You bring your uncertain data, and UADAPy wraps it in Distributions.

from uadapy import Distribution
import numpy as np

X, y = my_labeled_samples()
unique_y = np.unique(y)
grouped_X = [X[np.where(y == label)[0]] for label in unique_y]
# KDE of the underlying distributions (default when passing samples)
distributions_kde = [Distribution(grouped_X[i]) for i in range(len(unique_y))]

# or wrap a scipy.stats distribution for example
from scipy.stats import multivariate_normal
means = [np.mean(grouped_X[i], axis=0) for i in range(len(unique_y))]
covs = [np.cov(grouped_X[i], rowvar=False) for i in range(len(unique_y))]
distributions_gauss = [
  Distribution(multivariate_normal(mean=means[i], cov=covs[i], allow_singular=True)) for i in range(len(unique_y))
]

Then transform your distributions, using dimensionality reduction for instance.

# UAPCA readily projects gaussians
from uadapy.dr.uapca import uapca
gaussians_projected = uapca(distributions_gauss, n_dims=2)

# if distributions are not gaussians, GMMs can be leveraged
from uadapy.dr.wgmm_uapca import wgmm_uapca
from uadapy.distributions import multivariate_gmm
# KDE to GMM
gmms = [Distribution(multivariate_gmm.gmm_from_kde(d.kde)) for d in distributions_kde]
kdes_projected = wgmm_uapca(gmms, n_dims=2)

Visualize your distributions.

import matplotlib.pyplot as plt
from uadapy.plotting import plots_2d
fig, axs = plt.subplots(1, 1+len(unique_y), figsize=(2*len(unique_y), 2), sharex=True, sharey=True)
# combined plot of all distributions
plots_2d.plot_contour(kdes_projected, axs=axs[0], fig=fig)
axs[0].set_aspect('equal', adjustable='box')
# individual plots per distribution
for i in range(len(unique_y)):
    plots_2d.plot_contour(kdes_projected[i], axs=axs[i+1], fig=fig, distrib_colors=['#ff0088'])
    axs[i+1].set_aspect('equal', adjustable='box') 
plt.tight_layout()
plt.show()

Citation

If you use this software in your work, please cite it using the following metadata

@INPROCEEDINGS{UADAPy,
  author={Paetzold, Patrick and Hägele, David and Evers, Marina and Weiskopf, Daniel and Deussen, Oliver},
  booktitle={2024 IEEE Workshop on Uncertainty Visualization: Applications, Techniques, Software, and Decision Frameworks}, 
  title={UADAPy: An Uncertainty-Aware Visualization and Analysis Toolbox}, 
  year={2024},
  volume={},
  number={},
  pages={48-50},
  keywords={Uncertainty;Data analysis;Software packages;Conferences;Software algorithms;Pipelines;Data visualization;Python;Uncertainty visualization;software toolbox},
  doi={10.1109/UncertaintyVisualization63963.2024.00011}}