Forecasting Track Deep Dive¶

Projects:

projects/nyc-demand-forecasting-foundations-showcase
projects/demand-api-observability-showcase

Why This Deep Dive¶

Use this track when you want a time-aware forecasting workflow that extends into API serving and observability:

Train and evaluate forecasting models with chronological split discipline.
Serve predictions through an API with metrics and tracing hooks.

Phase 1: Forecasting Foundations¶

cd projects/nyc-demand-forecasting-foundations-showcase
make sync
make run
make verify

Quick demo mode:

make smoke

Key outputs:

artifacts/eval/metrics_summary.csv
artifacts/eval/prediction_examples.csv
artifacts/splits/time_split_manifest.json

Phase 2: Demand API Observability¶

cd projects/demand-api-observability-showcase
make sync
make train-demo
make test
make export-openapi
make verify
make dev

Key outputs:

artifacts/model.joblib
artifacts/metrics.json
openapi.json

Example: Inspect Forecast Metrics¶

cd projects/nyc-demand-forecasting-foundations-showcase
python - <<'PY'
import pandas as pd
df = pd.read_csv("artifacts/eval/metrics_summary.csv")
print(df.to_string(index=False))
PY

Example: Demand API Smoke Checks¶

curl -s http://127.0.0.1:8000/health
curl -s -X POST http://127.0.0.1:8000/predict \
  -H "content-type: application/json" \
  -d '{"pickup_zone_id":132,"pickup_datetime":"2026-02-13T09:00:00Z"}'
curl -s http://127.0.0.1:8000/metrics | head -n 10

For complete demand API request and response examples, see Demand API docs.

How To Interpret Outputs¶

time_split_manifest.json should confirm strict chronological separation.
Evaluate forecast quality with multiple metrics (MAE, RMSE, sMAPE) rather than one number.
Prediction API behavior should be checked alongside /metrics telemetry for operational readiness.
Contract checks should keep OpenAPI and runtime endpoint behavior aligned.

Next Step¶

Use Coverage Matrix to map this track to adjacent topics like drift monitoring, rollout decisions, and experiment tracking.