250
Nodes extracted from OSM. Intersections, not landmarks.
Compiling the hour
Compiling the hour
Network
Click a node. Watch attention light the rumor. This is not a street map for tourists. It is the adjacency the model believes — 250 intersections, 598 segments, Ember for the edge that matters.
Compiling attention graph
250
Nodes extracted from OSM. Intersections, not landmarks.
598
Directed-enough segments for industrial last mile and the highway pin.
5 min
The heartbeat. Coarser is policy. Finer is a sensor you do not have.
How to look
When you select a node, the scene does what GAT does: it raises the coefficient on some neighbors and lets others go dim. That is the opposite of a heat map that paints everything urgent. Urgency is a scarce color. We spent it on Ember.
Thiruvananthapuram is a capital hung on Palayam and the coast. The interesting failures are not the highway’s average speed. They are the industrial capillaries that inherit a jam twenty minutes late and keep it after the highway has forgotten. Spatial attention is how that delay becomes a first-class feature.
The public site cannot ship a 10-million-row tensor to a laptop fan. It can ship a faithful metaphor: nodes, edges, a slow orbit, a selected attention set. The scientific map — Folium, real geometry — still lives on the Flask runtime. Both are true. One is for the jury. One is for the GIS team.
A queue is a path. A gridlock is a cycle with too much flow. A flyover is a chord that sometimes lies about capacity. If your routing algorithm cannot see those as graph events, it is a travel agent. Traffic is not a travel agent.
Predicted speeds rewrite edge weights. PPO walks the same object you are orbiting. The orange vehicles on the home scene are the discrete version of that walk. If this feels like a lab instrument, good. It was supposed to.
Network is the abstract graph — nodes and attention as a teaching instrument. Map paints real OSM geometry with predicted speeds and routes. Influence zooms to one place and shows a heat map of road-to-road links. Use Network to understand the idea; Map to plan a trip; Influence to see who is coupled near Palayam or Technopark.
Those figures describe the educational subgraph used in early research storytelling. The live Flask map may load a larger Thiruvananthapuram edge file (tens of thousands of OSM segments). Both tell the same story: the city is a graph, not a single street.
This canvas emphasizes topology and attention. Speed colors and place-to-place routing live on Map after you generate a forecast. Speeds themselves are demo predictions keyed by time and scenario, on top of real OSM roads.
Controls
Drag to rotate. Scroll to zoom. The slow motion is intentional — a lab instrument, not a game.
Click an intersection. Ember highlights show which neighbors get attention in the metaphor.
Dim neighbors are still connected in the graph; they are simply not the loudest vote right now.
When you want real street geometry and speed colors, leave this metaphor and generate a Leaflet map.
Common questions
No. It is a teaching subgraph sized for the browser. Map/Flask may load a much larger edge CSV.
Attention is scarce on purpose. If everything glowed, nothing would mean influence.