pykale/pykale
Knowledge-Aware machine LEarning (KALE): accessible machine learning from multiple sources for interdisciplinary research, part of the 🔥PyTorch ecosystem. ⭐ Star to support our work!
What it solves
PyKale is designed to bridge the gap between data, software, and end users in machine learning, making it more accessible for interdisciplinary research. It specifically addresses the need for scalable, sustainable, and reusable workflows in multimodal learning and transfer learning, reducing redundancy and repetition in AI development.
How it works
PyKale uses a unified, pipeline-based API that organizes the machine learning process into distinct, modular stages:
- loaddata: Loads input data from disk or online sources.
- prepdata: Preprocesses and transforms data to fit machine learning modules.
- embed: Performs feature extraction or selection to learn new representations.
- predict: Generates the desired output predictions.
- evaluate: Measures performance using specific metrics.
- interpret: Analyzes and visualizes features and outputs for post-prediction analysis.
- pipeline: Combines these modules into a complete machine learning workflow.
Who it’s for
It is intended for both machine learning experts and interdisciplinary researchers (such as those in bioinformatics, graph analysis, and medical imaging) who need to build interpretable and knowledge-aware multimodal AI systems.
Highlights
- Multimodal Focus: Specialized for graphs, images, and videos.
- PyTorch Ecosystem: Officially listed in the the PyTorch Landscape under "Multimodal".
- pipeline-based API: Standardized, modular workflow for training and evaluation.
- Green ML Principles: Focuses on reducing, reusing, and recycling models and resources to increase efficiency.
- Interdisciplinary Application: Built for use in fields like medical imaging and bioinformatics.