dlordinal: a Python package for deep ordinal classification

TL;DR

The paper presents dlordinal, a PyTorch-based Python package that unifies state-of-the-art deep ordinal classification methods within a modular, extensible framework. It provides dedicated losses, output layers, soft labeling, and ordinal-aware metrics, along with dataset loaders for FGNet and Adience and Skorch integration for convenient experimentation. A comprehensive benchmark across three datasets evaluates multiple ordinal methods (e.g., Beta, Binomial, CLM, CLMWK, Exponential, Triangular, SB, OBD, Hybrid dropout), highlighting dataset- and metric-dependent performance and the trade-off between accuracy and efficiency. The results demonstrate the value of incorporating ordinal information and the practicality of a single framework to accelerate method development and benchmarking in real-world ordinal tasks. Overall, dlordinal offers a ready-to-use, open-source resource to advance research and applications in ordered classification problems across domains such as facial age estimation and medical imaging.

Abstract

dlordinal is a new Python library that unifies many recent deep ordinal classification methodologies available in the literature. Developed using PyTorch as underlying framework, it implements the top performing state-of-the-art deep learning techniques for ordinal classification problems. Ordinal approaches are designed to leverage the ordering information present in the target variable. Specifically, it includes loss functions, various output layers, dropout techniques, soft labelling methodologies, and other classification strategies, all of which are appropriately designed to incorporate the ordinal information. Furthermore, as the performance metrics to assess novel proposals in ordinal classification depend on the distance between target and predicted classes in the ordinal scale, suitable ordinal evaluation metrics are also included. dlordinal is distributed under the BSD-3-Clause license and is available at https://github.com/ayrna/dlordinal.

Figures (1)

• Figure 1: Representation of the classes and functions included in the various modules of the dlordinal package.