Repository navigation
新增
GBRBM 高斯-伯努利受限玻尔兹曼机;
MAIFS 模型无关特征选择(FeatureSelectionWrapper);
新增 Docker 开发环境;
用量统计。
改造
Q-VAE 重构为 AutoEncoderBase + config 驱动的工厂式构造;
Q-Diffusion 进行了接口调整,移除 EnergyBackboneAdapter Protocol;
dplm/qvae_mnist 示例由单体脚本拆分为 models/utils/workflows 分层结构。
文档
全量 rst 转 MyST Markdown,新增 10 篇理论基础、3 篇教程、贡献指南,数学渲染换 KaTeX。
New Features
Added GBRBM (Gaussian-Bernoulli Restricted Boltzmann Machine).
Added MAIFS (Model-Agnostic Feature Selection), integrating FeatureSelectionWrapper.
Added a new Docker development environment.
Added usage statistics tracking.
Refactoring
Refactored QVAE into an AutoEncoderBase + configuration-driven factory architecture.
Updated the QDiffusion interface and removed the EnergyBackboneAdapter Protocol.
Refactored the dplm/qvae_mnist example from a monolithic script into a layered structure with separate models, utils, and workflows modules.
Documentation
Migrated all documentation from reStructuredText (rst) to MyST Markdown.
Added 10 theoretical foundation documents, 3 tutorials, and a contribution guide.
Updated mathematical rendering to use KaTeX.