chap_core.hpo package

Submodules

chap_core.hpo.base module

class chap_core.hpo.base.Float(low: float, high: float, step: float | None = None, log: bool = False)

Bases: object

high: float

log: bool = False

low: float

step: float | None = None

class chap_core.hpo.base.Int(low: int, high: int, step: int = 1, log: bool = False)

Bases: object

high: int

log: bool = False

low: int

step: int = 1

chap_core.hpo.base.dedup(values)

Deduplicate while preserving order; works for scalars, lists, dicts, None.

chap_core.hpo.base.load_search_space_from_config(config: dict) → dict[str, Any]

chap_core.hpo.base.write_yaml(path: str, data: dict) → None

chap_core.hpo.cli module

chap_core.hpo.hpoModel module

class chap_core.hpo.hpoModel.HpoModel(searcher: Searcher, objective: Objective, direction: Literal['maximize', 'minimize'] = 'minimize', model_configuration: dict[str, list] | None = None)

Bases: HpoModelInterface

property get_best_config

get_leaderboard(dataset: Literal['hydro_met_subset', 'hydromet_clean', 'hydromet_10', 'hydromet_5_filtered', 'laos_full_data', 'uganda_data', 'ISIMIP_dengue_harmonized'] | None) → list[dict[str, Any]]

Runs hyperparameter optimization over the search space. Returns a sorted list of configurations together with their score.

predict(historic_data: DataSet, future_data: DataSet) → DataSet

train(dataset: Literal['hydro_met_subset', 'hydromet_clean', 'hydromet_10', 'hydromet_5_filtered', 'laos_full_data', 'uganda_data', 'ISIMIP_dengue_harmonized'] | None) → Tuple[str, dict[str, Any]]

Calls get_leaderboard to find the optimal configuration. Then trains the tuned model on the whole input dataset (train + validation).

write_best_config(output_yaml)

chap_core.hpo.hpoModelInterface module

class chap_core.hpo.hpoModelInterface.HpoModelInterface

Bases: ConfiguredModel

abstractmethod get_best_config() → dict

abstractmethod get_leaderboard(dataset: Literal['hydro_met_subset', 'hydromet_clean', 'hydromet_10', 'hydromet_5_filtered', 'laos_full_data', 'uganda_data', 'ISIMIP_dengue_harmonized'] | None) → list[dict[str, Any]]

chap_core.hpo.objective module

class chap_core.hpo.objective.Objective(model_template: ModelTemplate, metric: str = 'MSE', prediction_length: int = 3, n_splits: int = 4, ignore_environment: bool = False, debug: bool = False, log_file: str | None = None, run_directory_type: Literal['latest', 'timestamp', 'use_existing'] | None = 'timestamp')

Bases: object

chap_core.hpo.searcher module

class chap_core.hpo.searcher.GridSearcher

Bases: Searcher

ask() → dict[str, Any] | None

reset(search_space: dict[str, list]) → None

tell(params: dict[str, Any], result: float) → None

class chap_core.hpo.searcher.RandomSearcher(max_trials: int)

Bases: Searcher

Samples with replacement max_trials number of configurations.

ask() → dict[str, Any] | None

reset(search_space: dict[str, Any], seed: int | None = None) → None

tell(params: dict[str, Any], result: float) → None

class chap_core.hpo.searcher.Searcher

Bases: object

Abstract optimizer interface.

Implementations should: - call reset(space) before use - repeatedly return configurations via ask() until None (no more work) - receive feedback via tell(params, result)

ask() → dict[str, Any] | None

reset(space: Any) → None

tell(params: dict[str, Any], result: float) → None

class chap_core.hpo.searcher.TPESearcher(direction: str = 'minimize', max_trials: int | None = None)

Bases: Searcher

Tree Parzen Estimator. Parallel-safe TPE searcher using Optuna’s ask/tell with native distributions. - ask() returns a params dict that includes a reserved ‘_trial_id’. - tell() extracts ‘_trial_id’ from params to update the correct trial. Supports: - list[…] -> CategoricalDistribution - Float(low, high, step=None|>0, log=bool) -> FloatDistribution - Int(low, high, step>1, log=bool) -> IntDistribution

ask() → dict[str, Any] | None

reset(search_space: dict[str, list], seed: int | None = None) → None

tell(params: dict[str, Any], result: float) → None

chap_core.hpo.searcher.validate_search_space(search_space: dict[str, list])

Module contents