chap-models/chapkit_rwanda_malaria_bym_model

Rwanda Malaria BYM model (R-INLA), packaged as a chapkit shell-r service. # chapkit_rwanda_malaria_bym_model Rwanda Malaria BYM model (R-INLA), packaged as a chapkit shell-r service. A spatio-temporal Bayesian model for malaria incidence at the sector (ADM3) level, combining BYM spatial effects, RW1 temporal effects, and IID space-time interaction with lagged climate covariates. Implemented in R-INLA. The original MLproject lives at knutdrand/Kigali-Malaria-modelling-23-27-Feb; this repo wraps the same R-INLA model body in chapkit's HTTP service so it can be discovered, run, and managed by chap-core like any other registered ML service. ## Layout ``` chapkit_rwanda_malaria_bym_model/ ├── main.py # ShellModelRunner wiring + MLServiceInfo metadata ├── scripts/ │ ├── train.R # no-op (BYM fits during predict) │ └── predict.R # BYM + RW1 + IID, R-INLA, requires polygons ├── pyproject.toml ├── uv.lock ├── Dockerfile # FROM ghcr.io ├── compose.yml └── Makefile ``` ## Quick start ```bash uv lock docker compose up --build # In another shell: uv run chapkit test --url localhost ``` `make test` builds the image, boots a one-shot container, runs `chapkit test` against it (which synthesises panel data + polygons and drives a full train -> predict cycle), and tears the container down. ## Model details - **Family**: Poisson with `offset = log(population)`. - **Spatial**: BYM (`f(ID, model = "bym", graph = ...)`), with a rook adjacency graph built at predict-time from the supplied polygons. - **Temporal**: RW1 over month-year time index. - **Space-time interaction**: IID over (sector x time). - **Covariates**: lagged climate variables (configurable depth, default 2). - **Required covariates**: population, rainfall, mean_temperature, relative_humidity. Free additional continuous covariates are accepted via the chap-core config. ## Configuration `main.py` exposes a small Config class that translates to YAML in the workspace at predict-time: | Field …