Spaces:
Paused
Paused
| title: DesirePath | |
| emoji: 🌡️ | |
| colorFrom: blue | |
| colorTo: yellow | |
| sdk: docker | |
| app_port: 7860 | |
| pinned: false | |
| license: mit | |
| short_description: What NYC's quarter-mile cooling promise delivers | |
| # DesirePath | |
| **The contribution is the OpenSidewalks graph. This is a consumer of it.** | |
| New York's Cool It! plan claims that "no New Yorker in the most heat-burdened | |
| communities is more than 1/4 mile away from an outdoor cooling element" | |
| (NYC DEP, 24 June 2020). That quarter mile is a straight line. DesirePath | |
| measures the same claim along the sidewalk network a person actually has to | |
| walk, with heat priced into the cost of every metre. | |
| In Brownsville the claim counts 12.8% of the sidewalk network as covered. | |
| Walking it in the heat reaches 3.3%. North Corona has no cooling element at all. | |
| Two screens: `/coverage` measures the claim, `/` plans a walk from a typed | |
| sentence. | |
| ## How heat enters the routing | |
| Every edge carries its own mean radiant temperature, the number that captures | |
| the radiant heat load a body actually exchanges with a street. In this graph it | |
| spans 32.5 °C in deep shade to 70.0 °C in open sun. The cost Dijkstra minimises: | |
| ``` | |
| cost = length x (accessibility multipliers) x (1 + sun_inflation x exposure) | |
| exposure = clamp((mrt_edge - 33.01) / (61.8 - 33.01), 0, 1) | |
| ``` | |
| `sun_inflation` runs 0.16 to 0.84, from Melnikov et al. (2022), estimated from | |
| 408 observed pedestrian path choices. At the top of that range a metre in full | |
| sun is charged as 1.84 metres, so the router accepts a detour to avoid it. | |
| Adding this to the accessibility router needed one new construct and no new | |
| concepts: heat became one more edge attribute beside kerb height, and | |
| sensitivity to it one more term in the same cost tree. | |
| Mean radiant temperature here is a **proxy, not SOLWEIG**. Thermal survey covers | |
| five neighbourhoods, 7.67% of the city's edges. | |
| ## Where the model runs | |
| This Space runs Granite 4 itself, so the first query is fast and nothing is | |
| downloaded. **Your typed sentence is sent to this Space to be read into a form.** | |
| Append `?llm=webgpu` to run the model in your own browser instead. It downloads | |
| once (about 30 seconds, cached afterwards) and the sentence never leaves the | |
| page. That path needs Chrome or Edge with WebGPU. | |
| Routing, the graph and the geocoder are always local to your browser either way. | |
| Source: https://github.com/msradam/desirepath-app | |