kats.models.metalearner.get_metadata module¶

A module for computing the meta-data of time series.

This module contains the class for computing the meta-data of time series. The meta-data of a time series is consists of three parts: 1) time series features; 2) the best hyper-parameters of each candidate models and their corresponding errors; and 3) the best model.

class kats.models.metalearner.get_metadata.GetMetaData(data: kats.consts.TimeSeriesData, all_models: Dict[str, Any] = {'arima': <class 'kats.models.arima.ARIMAModel'>, 'holtwinters': <class 'kats.models.holtwinters.HoltWintersModel'>, 'prophet': <class 'kats.models.prophet.ProphetModel'>, 'sarima': <class 'kats.models.sarima.SARIMAModel'>, 'stlf': <class 'kats.models.stlf.STLFModel'>, 'theta': <class 'kats.models.theta.ThetaModel'>}, all_params: Dict[str, Any] = {'arima': <class 'kats.models.arima.ARIMAParams'>, 'holtwinters': <class 'kats.models.holtwinters.HoltWintersParams'>, 'prophet': <class 'kats.models.prophet.ProphetParams'>, 'sarima': <class 'kats.models.sarima.SARIMAParams'>, 'stlf': <class 'kats.models.stlf.STLFParams'>, 'theta': <class 'kats.models.theta.ThetaParams'>}, min_length: int = 30, scale: bool = True, method: kats.consts.SearchMethodEnum = <SearchMethodEnum.RANDOM_SEARCH_UNIFORM: 2>, executor: Optional[Any] = None, error_method: str = 'mae', num_trials: int = 5, num_arms: int = 4, **kwargs)[source]¶

Bases: object

A class for generate meta-data of time series.

The meta-data of time series contains three components:

  1. Time series feature vector, such as entropy, seasonal features, ACF/PACF based features.

  2. Best hyper-parameters for each candidate models and their corresponding errors for a time series data.

  3. Best model for a time series data.

This class provides tune_executor and get_meta_dat.

data¶

kats.consts.TimeSeriesData object representing the input time series data.

all_models¶

Optional; A dictionary of candidate model classes. Default dictionary includes models of ARIMA, SARIMA, HoltWinters, Prophet, Theta, and STLF.

all_params¶

Optional; A dictionary of the corresponding candidate model parameter classes. Default includes model parameter classes of ARIMA, SARIMA, HoltWinters, Prophet, Theta, and STLF.

min_length¶

Optional; An integer for the minimal length of a time series. Time series data whose length is shorter than min_length will be excluded. Default is 30.

scale¶

Optional; A boolean to specify whether or not to rescale the time series data by its maximum values. Default is True. Default is True.

method¶

Optional; A SearchMethodEnum object defining the search method for hyper-parameters tuning. Default is random search in the default parameter space.

executor¶

Optional; A callable parallel executor for tuning individual candidate models in parallel. Default is the native implementation with Python’s multiprocessing.

error_method¶

Optional; A string for error metric used for model evaluation. Can be ‘mape’, ‘smape’, ‘mae’, ‘mase’, ‘mse’, or ‘rmse’. Default is ‘mae’.

num_trials¶

Optional; An integer for the number of trials in hyper-parameter search. Default is 5.

num_arms¶

Optional; An integer for the number of arms in hyper-parameter search. Default is 4.

Sample Usage:
>>> TSdata = TimeSeriesData(data)
>>> MD = GetMetaData(data=TSdata)
>>> hpt_res = MD.tune_executor()
>>> my_meta_data = MD.get_meta_data() # Get meta-data, hyper-parameter searching method and error metric.
get_meta_data() → Dict[str, Any][source]¶

Get meta data, as well as search method and type of error metric

Meta data includes time series features, best hyper-params for each candidate models, and best model.

Returns

A dictionary storing the best hyper-parameters and the errors for each candidate model, the features of the time series data, the hyper-parameter searching method, the error metric used for model evaluation and the corresponding best model.

tune_executor() → Dict[str, Any][source]¶

Get the best hyper parameters for each candidate model and their corresponding errors for the time series data.

Returns

A dictionary storing the best hyper-parameters and the errors for each candidate model.