Skip to content

Evaluation & prediction

This is step 6 of the workshop. The configuration on this page is the single reference used by both walkthroughs, so switch between them without re-entering a different scenario.

With DHIS2, chap-core, and the apps running, you can now put a model to work. There are two things you typically do with a model:

  • Evaluate (backtest) - run the model over historical data and compare its predictions to what actually happened, to see how well it performs.
  • Predict - run the model on the latest data to forecast the coming periods.

Choose how you want to drive the same workflow:

Both use the same configuration, described below. Because everyone is working from the same Laos climate demo database, you can copy these exact values.

The model
Model CHAP-EWARS Model (chapkit)
Configured model chapkit-ewars-model - always refer to it by this canonical name; the numeric id (e.g. 12) is assigned by your database and differs per install
What it is A Bayesian hierarchical model (WHO EWARS) implemented with INLA, packaged as a chapkit model.
Target disease_cases
Required covariate population
Extra covariates rainfall, mean_temperature
Period type Monthly

The data (Laos demo)

We model monthly dengue cases across the 18 provinces of Lao PDR, using climate covariates. The model's features map to these DHIS2 data items:

Model feature DHIS2 data item dataElement id
Disease cases (target) Dengue Cases (Any) - Weekly SK9a8nJJTAI
Population LSB: Population (Estimated-single age) D8Q6nNeQ7i3
Rainfall CCH - Precipitation (CHIRPS) DZte8CXJ6zJ
Mean temperature CCH - Air temperature (ERA5-Land) Pjd8Rn6mTb0
Organisation units Province level (level 2) - all 18 provinces
Period range 2023-01 to 2024-12 (Monthly)
Prediction horizon 3 months (the default n_periods; forecasts 2025-01 to 2025-03)
The 18 province org-unit IDs 
W6sNfkJcXGC  01 Vientiane Capital   YvLOmtTQD6b  02 Phongsali
XKGgynPS1WZ  03 Louangnamtha        rO2RVJWHpCe  04 Oudomxai
FRmrFTE63D0  05 Bokeo               MBZYTqkEgwf  06 Louangphabang
hdeC7uX9Cko  07 Houaphan            RdNV4tTRNEo  08 Xainyabouli
VWGSudnonm5  09 Xiangkhouang        quFXhkOJGB4  10 Vientiane
vBWtCmNNnCG  11 Bolikhamxai         c4HrGRJoarj  12 Khammouan
pFCZqWnXtoU  13 Savannakhet         TOgZ99Jv0bN  14 Salavan
dOhqCNenSjS  15 Xekong              sv6c7CpPcrc  16 Champasak
hRQsZhmvqgS  17 Attapu              K27JzTKmBKh  18 Xaisomboun

How DHIS2, the app, and chap fit together

The Modelling App (and your curl commands) reach chap-core through the DHIS2 route you set up earlier. The chap API lives under:

http://localhost:8080/api/routes/chap/run/v1/...

with the pieces you will use:

  • …/v1/crud/configured-models - the models, e.g. id 12
  • …/v1/crud/backtests - evaluations
  • …/v1/crud/prediction-setups - reusable prediction configs (created from a backtest, then run)
  • …/v1/crud/predictions - the forecasts a setup run produces
  • …/v1/jobs - the running/finished jobs

Before you start

DHIS2 + chap-core are running and connected (Connect CHAP), and the Modelling App is installed (Install the apps).

Choose a walkthrough

Start with the Modelling App walkthrough for the main workshop path. Use the API walkthrough when you want to understand or automate the same requests.