Semi-supervised classification leverages both labeled and unlabeled data to improve predictive performance, but existing software support is fragmented across methods and modalities. We introduce ModSSC, an open source Python framework that unifies inductive and transductive semi-supervised classification in a modular code base. ModSSC implements a broad range of classical and recent algorithms, provides loaders for tabular, image, text, audio and graph datasets, and exposes a single configuration interface for specifying datasets, models and evaluation protocols. It supports both lightweight classical methods on small datasets running on CPU and recent deep approaches that can exploit multiple GPUs within the same experimental framework. Experiments are described declaratively in YAML, which facilitates reproducing existing work and running large comparative studies. ModSSC 1.0.0 is released under the MIT license with extensive documentation and tests, and is available at https://github.com/ModSSC/ModSSC.

翻译：半监督分类通过同时利用标注和未标注数据来提升预测性能，但现有软件支持在不同方法与模态间较为分散。本文提出ModSSC，一个开源的Python框架，以模块化代码库统一了归纳式与直推式半监督分类。ModSSC实现了广泛的经典与前沿算法，提供了针对表格、图像、文本、音频及图数据集的加载器，并通过单一配置接口支持数据集、模型与评估协议的指定。该框架既支持在CPU上运行、面向小数据集的轻量级经典方法，也支持在同一实验框架下利用多GPU的前沿深度学习方法。实验可通过YAML文件以声明式方式描述，便于复现现有工作与开展大规模对比研究。ModSSC 1.0.0版本采用MIT许可证发布，提供详尽的文档与测试，可通过https://github.com/ModSSC/ModSSC获取。