Workshop 08: Spatial Intelligence for Architecture: Graph Analytics and Graph Machine Learning with TopologicPy
Workshop providers:
Prof. Wassim Jabi, Welsh School of Architecture, JabiW@cardiff.ac.uk
Dania Al-Harasis, Welsh School of Architecture, Al-HarasisD@cardiff.ac.uk
Description:
Buildings and cities are increasingly understood not merely as collections of geometric objects but as complex spatial networks whose structure influences accessibility, movement, environmental performance, health outcomes, social interaction, and operational efficiency. Recent advances in graph theory, network science, and graph machine learning have created new opportunities for analysing, understanding, and learning from these spatial structures at scales ranging from individual rooms to entire urban districts.
This intensive two-day workshop introduces participants to the emerging field of Spatial Intelligence using TopologicPy, an open-source Python library that integrates geometry, topology, graph theory, semantics, and machine learning for Architecture, Engineering, Construction, and Operations (AECO). The workshop provides both theoretical foundations and hands-on experience in representing buildings as graphs and applying advanced analytical and machine learning techniques to spatial data.
During the first day, participants will learn how architectural and BIM models can be transformed into topological and graph-based representations. Topics include adjacency, circulation, visibility, connectivity, and semantic networks, as well as the computation of graph metrics that support the analysis of spatial performance, accessibility, resilience, and building functionality. Participants will construct and analyse spatial graphs derived from architectural models using TopologicPy.
The second day focuses on Graph Machine Learning and its application to the built environment. Participants will learn how to prepare graph datasets, engineer node and edge features, and train Graph Neural Networks (GNNs) for tasks such as space classification, building performance prediction, and pattern recognition. The workshop will introduce contemporary approaches including GraphSAGE, graph embeddings, semantic enrichment, and emerging GraphRAG workflows that combine graph databases and large language models.
Through a series of guided exercises, participants will develop complete workflows that move from geometric and BIM data to graph-based analysis and machine learning. Participants are encouraged to bring their own floor plans, BIM models, or spatial datasets, allowing them to apply the techniques directly to their own research and design projects.
Day 1: Building Spatial Graphs and Spatial Analytics
09:00 – 09:30 Welcome, Introduction, and Installation Check
09:30 – 10:30 From Geometry to Topology: Spatial Modelling with TopologicPy
10:30 – 10:45 Break – Debugging Installation Issues
10:45 – 13:00 Graph-Based Representation of Buildings: Adjacency, Circulation, Visibility, and Connectivity Networks
13:00 – 14:00 Lunch
14:00 – 15:00 Deriving Graphs from IFC Models: Semantic and Spatial Graphs
15:00 – 15:30 Break – One-on-one consultations
15:30 – 17:00 Spatial Intelligence: Graph-theoretic Metrics
Day 2: Graph Machine Learning and Spatial Intelligence
09:00 – 09:30 Recap and Introduction to Graph Machine Learning
09:30 – 11:30 Preparing Architectural Graph Datasets for Machine Learning
11:30 – 12:00 Break – One-on-One consultancies
12:00 – 13:00 Graph Neural Networks for Architecture
13:00 – 14:00 Lunch
14:00 – 15:00 Hands-On Training of Graph Machine Learning Models Using TopologicPy and PyTorch Geometric
15:00 – 16:00 Interpreting Results: Prediction, Classification, and Explainability
16:00 – 17:00 Closing Discussion – Preparation of output
Software Installations
Participants should install prior to the workshop:
- Python 3.11 or later
- Jupyter Notebook or JupyterLab
- TopologicPy
- PyTorch
- PyTorch Geometric
- Plotly
A preconfigured installation package and workshop notebooks will be provided before the event.
Workshop type:
2 day online only workshop
