API

hcrystalball.model_selection Package

Functions

add_model_to_gridsearch(model, grid_search)	Extends gridsearch with provided model.
filter_data(df[, include_rules, exclude_rules])	Filter provided dataframe by {column:value} rules.
get_gridsearch(frequency[, horizon, …])	Get grid search object based on selection criteria.
load_model_selector(folder_path)	Load information about stored model selection
partition_data(df, partition_by)	Partition data by values found in one or more columns.
partition_data_by_values(df, column, …[, …])	Partition data by one column and a fixed set ov values within that column.
prepare_data_for_training(df, frequency, …)	Prepare data for model selection.
run_model_selection(df, grid_search, …[, …])	Run parallel cross validation on data and select best model
select_model(df, target_col_name, …[, …])	Find the best model from grid_search for each partition
select_model_general(df, grid_search, …[, …])	Run cross validation on data and select best model
get_best_not_failing_model(grid_search, X, y)	Prevent situation when model incompatible data are not seen during CV (in the last split of actuals) and model which cannot be fitted on the full dataset is choseen.

Classes

FinerTimeSplit([n_splits, horizon, …])	Time series cross-validator.
ModelSelector(horizon, frequency[, …])	Enable large scale cross validation easily accessible.
ModelSelectorResult(best_model, cv_results, …)	Consolidate infromation/methods from cross validation for 1 time series

Variables

load_best_model(*[, expert_type, …])	Returns unpickled file or json metadata from directory.
load_model_selector_result(*[, expert_type, …])	Returns unpickled file or json metadata from directory.

hcrystalball.wrappers Package

Functions

get_sklearn_wrapper(model_cls, **model_params)

Factory function returning the model specific SklearnWrapper with provided model_cls parameters.

Classes

ProphetWrapper([growth, changepoints, …])	Wrapper for prophet.Prophet model
ExponentialSmoothingWrapper([endog, trend, …])	Wrapper for ExponentialSmoothing (see other parameters there)
SimpleSmoothingWrapper([endog, …])	Wrapper for SimpleExpSmoothing (see other parameters there)
HoltSmoothingWrapper([endog, exponential, …])	Wrapper for Holt (see other parameters there)
ThetaWrapper([endog, period, deseasonalize, …])	Wrapper for ThetaModel (see other parameters there)
SarimaxWrapper([order, seasonal_order, …])	Wrapper for ARIMA and AutoARIMA
TBATSWrapper([use_box_cox, box_cox_bounds, …])	Wrapper for TBATS model
BATSWrapper([use_box_cox, box_cox_bounds, …])	Wrapper for BATS model

hcrystalball.ensemble Package

Classes

SimpleEnsemble(base_learners[, …])	SimpleEnsemble model, which takes a list of any hcrystalball model wrapper instance(s) as base learners and aggregates their prediction using ensemble_func.
StackingEnsemble(base_learners, meta_model)	StackingEnsemble model, which takes a list of any hcrystalball model wrapper instance(s) as base learners.

hcrystalball.feature_extraction Package

Classes

HolidayTransformer([country_code, …])	Generate holiday feature based on provided ISO code
SeasonalityTransformer([auto, freq, …])	Generate seasonal feature columns using one-hot encoding.

hcrystalball.compose Package

Classes

TSColumnTransformer(transformers, *[, …])

Time Series compatible ColumnTransformer.

hcrystalball.preprocessing Package

Classes

TargetTransformer(estimator, y_transformer)

Enable transformation of the target.

hcrystalball.metrics Package

Functions

get_scorer([function])	Get a scorer supporting storing data for gridsearch from string.
make_ts_scorer(score_func[, …])	Make a scorer from a performance metric or loss function.

hcrystalball.utils Module

Functions

get_estimator_repr(model[, n_char_max])	Get the string representation of a model
get_sales_data([n_dates, n_assortments, …])	Load subset of Rossmann store sales dataset.
generate_estimator_hash(model)	Generate a unique hash for a model using its string representation
generate_multiple_tsdata([n_dates, …])	Provide easy way how to generate dummy data for tests or tutorial purposes
generate_partition_hash(partition_label)	Generate a unique hash for data partition
generate_tsdata([n_dates, random_state])	Generate dummy daily time series data compatible with hcrystalball API.
optional_import(module_path, class_name, …)	Imports optional dependencies.
set_verbosity(func)