CartoBoost ServiceTimeResidualModel

Use ServiceTimeResidualModel when you already have a required baseline numeric estimate and want CartoBoost to learn the residual correction. The final prediction is the baseline plus the learned residual.

Python Example

from cartoboost.deep import ServiceTimeResidualModel

rows = [
    {
        "baseline_value": 12.0,
        "actual_value": 13.5,
        "features": [0.2, 1.0, 4.0],
    },
    {
        "baseline_value": 9.5,
        "actual_value": 8.9,
        "features": [0.1, 0.0, 3.0],
    },
]

model = ServiceTimeResidualModel()
model.fit(rows)
prediction = model.predict(rows, return_interval=True)
model.save("service-residual.json")

Browser WASM Example

ServiceTimeResidualModel Wasm example

Runs the Rust-backed browser export on a substantial multi-route taxi panel with a held-out prediction window.

Ready to run in this page.

When To Use

• A baseline estimate is required by the workflow.
• The model should correct residual error rather than replace the baseline.
• Features explain systematic bias around the baseline.
• Missing baseline values should be treated as data errors.

Use When

Need	Better first choice
Correct a known numeric baseline.	ServiceTimeResidualModel
Fit a numeric model from raw features only.	CartoBoostRegressor
Calibrate uncertainty around predictions.	Probabilistic and conformal models
Forecast future time-indexed values.	Forecasting model guides

Validation

Compare the corrected prediction against the baseline alone on the same split. Report residual MAE/RMSE and final prediction MAE/RMSE so readers can see whether the correction is useful or just adding variance.

Limitations

• The baseline estimate is required and its errors bound final performance.
• Residual correction can amplify noise when baseline error has little predictable structure.
• Prediction intervals require separate calibration and validation.