ThymeBoost package

Subpackages

• Exogenous Models
  • Decision Tree
  • GLM
  • OLS
• ThymeBoost.fit_components package
  • Submodules
  • ThymeBoost.fit_components.fit_exogenous module
  • ThymeBoost.fit_components.fit_seasonality module
    • FitSeasonality
      • FitSeasonality.fit_seasonal_component()
      • FitSeasonality.set_estimator()
  • ThymeBoost.fit_components.fit_trend module
  • Module contents
• ThymeBoost.fitter package
  • Submodules
  • ThymeBoost.fitter.booster module
  • ThymeBoost.fitter.decompose module
  • Module contents
• Seasonality Models
  • Classic
    • ClassicSeasonalityModel
      • ClassicSeasonalityModel.fit()
      • ClassicSeasonalityModel.model
      • ClassicSeasonalityModel.predict()
  • Fourier
    • FourierSeasonalityModel
      • FourierSeasonalityModel.fit()
      • FourierSeasonalityModel.get_fourier_series()
      • FourierSeasonalityModel.handle_seasonal_weights()
      • FourierSeasonalityModel.model
      • FourierSeasonalityModel.predict()
• Trend Models
  • ARIMA
  • Smoothers
  • EWM
    • EwmModel
      • EwmModel.fit()
      • EwmModel.model
      • EwmModel.predict()
  • Linear
  • LOESS
    • LoessModel
      • LoessModel.fit()
      • LoessModel.model
      • LoessModel.predict()
  • Mean
    • MeanModel
      • MeanModel.fit()
      • MeanModel.model
      • MeanModel.predict()
  • Median
    • MedianModel
      • MedianModel.fit()
      • MedianModel.model
      • MedianModel.predict()
  • RANSAC
• Utilities
  • ThymeBoost.utils.plotting module
    • plot_components()
    • plot_optimization()
    • plot_results()
  • ThymeBoost.utils.trend_dampen module
    • trend_dampen()

Submodules

ThymeBoost.ThymeBoost module

ThymeBoost.cost_functions module

ThymeBoost.cost_functions.calc_cost(time_series, prediction, c, regularization, global_cost)

ThymeBoost.cost_functions.calc_entropy(actuals, predicted)

ThymeBoost.cost_functions.calc_mae(actuals, predicted)

ThymeBoost.cost_functions.calc_mape(actuals, predicted, epsilon=1e-08)

ThymeBoost.cost_functions.calc_mse(actuals, predicted)

ThymeBoost.cost_functions.calc_smape(actuals, predicted)

ThymeBoost.cost_functions.get_split_cost(time_series, split1, split2, split_cost)

Calculate the cost for the given split point using binary segmentation.

Parameters:

• time_series (np.array) – The input time series.

• split1 (np.array) – The first split of the binary segmentation.

• split2 (np.array) – The second split of the binary segmentation..

• split_cost (str) – The metric to use for calculating the cost.

Raises:

ValueError – If the cost metric provided does not exist this error will be raised.

Returns:

cost – The cost for the given split.

Return type:

float

ThymeBoost.ensemble module

ThymeBoost.optimizer module

ThymeBoost.param_iterator module

A base class which is inherited by both ensemble and optimize classes. Used to clean large parameter lists of illegal combinations

class ThymeBoost.param_iterator.ParamIterator

Bases: object

The ensemble/optimizer base class

param_check(params)

Given a dict of params, check for illegal combinations

Parameters:

params (dict) – A dictionary of params for one single thymeboost model.

Returns:

params – A dictionary with illegal values nullified.

Return type:

dict

sanitize_params(param_list)

Iterate through param dicts to sanitize illegal combinations.

Parameters:

param_list (list) – A List of param dicts.

Returns:

List of cleaned param dicts.

Return type:

list

ThymeBoost.predict_functions module

ThymeBoost.predict_functions.predict_exogenous(booster_obj, future_exo, boosting_round, forecast_horizon)

Predict the exogenous component using the booster.

Parameters:

• boosting_round (int) – The round to reference when getting model params.

• forecast_horizon (int) – Number of periods to forecast.

Returns:

seas_round – That boosting round’s predicted seasonal component.

Return type:

np.array

ThymeBoost.predict_functions.predict_rounds(booster_obj, forecast_horizon, trend_penalty, future_exo=None, online_learning=False)

Predict all the rounds from a booster

Parameters:

• fitted_output (pd.DataFrame) – Output from fit method.

• forecast_horizon (int) – Number of periods to forecast.

Returns:

• trend_predictions (np.array) – Trend component.

• seasonal_predictions (np.array) – seasonal component.

• predictions (np.array) – Predictions.

ThymeBoost.predict_functions.predict_seasonality(booster_obj, boosting_round, forecast_horizon)

Predict the seasonality component using the booster.

Parameters:

• boosting_round (int) – The round to reference when getting model params.

• forecast_horizon (int) – Number of periods to forecast.

Returns:

seas_round – That boosting round’s predicted seasonal component.

Return type:

np.array

ThymeBoost.predict_functions.predict_trend(booster_obj, boosting_round, forecast_horizon, trend_penalty, online_learning)

Predict the trend component using the booster

Parameters:

• boosting_round (int) – The round to reference when getting model params.

• forecast_horizon (int) – Number of periods to forecast.

Returns:

trend_round – That boosting round’s predicted trend component.

Return type:

np.array

ThymeBoost.split_proposals module

A class to propose splits according to the most and least interesting points based on the gradient.

class ThymeBoost.split_proposals.SplitProposals(given_splits, exclude_splits, min_sample_pct, n_split_proposals, split_strategy, approximate_splits)

Bases: object

Generate splits to try when fit_type = ‘local’.

Parameters:

• given_splits (list) – Splits to use when using fit_type=’local’.

• exclude_splits (list) – exclude these index when considering splits for fit_type=’local’. Must be idx not datetimeindex if using a Pandas Series.

• min_sample_pct (float) – Percentage of samples required to consider a split. Must be 0<min_sample_pct<=1.

• n_split_proposals (int) – Number of split proposals based on the gradients.

• approximate_splits (boolean) – Whether to use proposal splits based on gradients or exhaustively try splits with at least min_sample_pct samples.

Return type:

A list of splits to try.

get_approximate_split_proposals()

Propose indices to split the data on.

Returns:

proposals – The proposals to try out.

Return type:

list

get_split_proposals(time_series)

Get proposals based on simple gradient strategy or histograms.

Parameters:

time_series (np.array) – The time series.

Returns:

A list of split indices to try.

Return type:

list

gradient_based_proposal()

histogram_based_proposal()

set_proposal_method()

Module contents