Predict module

class pyro_risks.pipeline.predict.PyroRisk(model: Optional[str] = 'RF')[source]

Bases: object

Pyronear Wildfire Risk Forecaster

Load a trained pipeline from pyrorisks remote model registry, download features from publicly available data sources (CDS API). Forecast the local (NUTS 3 level) daily wildfire risks (forest fire danger) in a Given Country (France).

Parameters

model – Can be ‘RF’ for random forest or ‘XGBOOST’ for xgboost. Defaults to ‘RF’.

Raises

ValueError – Model can be only of type RF or XGBOOST

expose_predictions(day: str, country: Optional[str] = 'France', dir_path: Optional[str] = None, dir_destination: Optional[str] = None) → dict[source]

Serves a prediction for the specified day.

Parameters

  • day – Date of interest (‘%Y-%m-%d’) for example ‘2020-05-05’.

  • country – Country of interest. Defaults to ‘France’.

  • dir_path – Location of the predictions to download, relative to the root of the dvc project.

  • to None. (Defaults) –

Returns

keys are departements, values dictionaries whose keys are score and explainability and values probability predictions for label 1 (fire) and feature contributions to predictions respectively.

Return type

dict[dict]

static get_inputs(day: str, country: Optional[str] = 'France', dir_destination: Optional[str] = None) → None[source]

Download datasets and build features for forecasting daily wildfire risks on a given date.

The get_inputs method downloads datsets from publicly available data sources (CDS API) and build features for forecasting wildfire risks on a given date. The downloaded inputs are persited in the destination csv file.

Parameters

  • day – Date of interest (‘%Y-%m-%d’) for example ‘2020-05-05’.

  • country – Country of interest. Defaults to ‘France’.

  • destination – Location where the daily inputs are persisted.

  • to None (Defaults) –

get_pipeline(path: Optional[str] = None, destination: Optional[str] = None) → None[source]

Download trained pipeline from remote model registry.

The get_pipeline method downloads the selected trained pipeline from the pyrorisks remote model registry. The downloaded pipeline is persited in the destination joblib file.

Parameters

  • path – Location and file name of the pipeline to download, relative to the root of the

  • project. Defaults to None (dvc) –

  • destination – Location where the pipeline is downloaded. Defaults to None (self.model_path).

get_predictions(day: str, country: Optional[str] = 'France', dir_path: Optional[str] = None, dir_destination: Optional[str] = None) → None[source]

Download predictions for the day of interest from the remote prediction registry.

The get_predictions method downloads the forecasted local wildfire risks at a given day in a given country (France). The downloaded predictions are persited in the destination joblib file.

Parameters

  • day – Date of interest (‘%Y-%m-%d’) for example ‘2020-05-05’.

  • country – Country of interest. Defaults to ‘France’.

  • dir_path – Location of the predictions to download, relative to the root of the dvc project.

  • to None. (Defaults) –

  • dir_destination – Location where the daily predictions are persisted. Defaults to None.

load_inputs(day: str, country: Optional[str] = 'France', usecols: Optional[List[str]] = None, dir_path: Optional[str] = None) → None[source]

Load inputs from local path.

Parameters

  • day – Date of interest (‘%Y-%m-%d’) for example ‘2020-05-05’.

  • country – Country of interest. Defaults to ‘France’.

  • dir_path – Location where the daily inputs have been persisted. Defaults to None

  • (cfg.PREDICTIONS_REGISTRY). –

load_pipeline(path: Optional[str] = None) → None[source]

Load trained pipeline from local path.

Parameters

path – Location where the pipeline has been downloaded. Defaults to None (self.model_path).

predict(day: str, country: Optional[str] = 'France', zone_column: Optional[str] = 'departement', dir_destination: Optional[str] = None) → None[source]

Predict local daily wildfire risks in a given country.

Forecast the local (NUTS 3 level) daily wildfire risks (forest fire danger) in a given country (France). Note that predictions on fwi and era5land data queried from CDS API will return 93 departments instead of 96 for France.

Parameters

  • day – Date of interest (‘%Y-%m-%d’) for example ‘2020-05-05’.

  • country – Country of interest. Defaults to ‘France’.

  • dir_destination – Location where the daily inputs are persisted.

  • to None (Defaults) –