pmdarima.model_selection.RollingForecastCV

class pmdarima.model_selection.RollingForecastCV(h=1, step=1, initial=None)[source][source]

Use a rolling forecast to perform cross validation

Sometimes called “evaluation on a rolling forecasting origin” [1], this approach to CV incrementally grows the training size while using a single future sample as a test sample, e.g.:

With h == 1:

array([15136., 16733., 20016., 17708., 18019., 19227., 22893., 23739.])
1st: ~~~~ tr ~~~~ tr ~~~~ te
2nd: ~~~~ tr ~~~~ tr ~~~~ tr ~~~~ te
3rd: ~~~~ tr ~~~~ tr ~~~~ tr ~~~~ tr ~~~~ te

With h == 2:

array([15136., 16733., 20016., 17708., 18019., 19227., 22893., 23739.])
1st: ~~~~ tr ~~~~ tr ~~~~ te ~~~~ te
2nd: ~~~~ tr ~~~~ tr ~~~~ tr ~~~~ te ~~~~ te
3rd: ~~~~ tr ~~~~ tr ~~~~ tr ~~~~ tr ~~~~ te ~~~~ te
Parameters:

h : int, optional (default=1)

The forecasting horizon, or the number of steps into the future after the last training sample for the test set.

step : int, optional (default=1)

The size of step taken to increase the training sample size.

initial : int, optional (default=None)

The initial training size. If None, will use 1 // 3 the length of the time series.

Attributes

horizon The forecast horizon for the cross-validator

References

[R86]https://robjhyndman.com/hyndsight/tscv/

Examples

With a step size of one and a forecasting horizon of one, the training size will grow by 1 for each step, and the test index will be 1 + the last training index:

>>> import pmdarima as pm
>>> from pmdarima.model_selection import RollingForecastCV
>>> wineind = pm.datasets.load_wineind()
>>> cv = RollingForecastCV()
>>> cv_generator = cv.split(wineind)
>>> next(cv_generator)
(array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
        17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
        34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50,
        51, 52, 53, 54, 55, 56, 57]), array([58]))
>>> next(cv_generator)
(array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
        17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
        34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50,
        51, 52, 53, 54, 55, 56, 57, 58]), array([59]))

With a step size of 2 and a forecasting horizon of 4, the training size will grow by 2 for each step, and the test index will 4 + the last index in the training fold:

>>> cv = RollingForecastCV(step=2, h=4)
>>> cv_generator = cv.split(wineind)
>>> next(cv_generator)
(array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
       17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
       34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50,
       51, 52, 53, 54, 55, 56, 57]), array([58, 59, 60, 61]))
>>> next(cv_generator)
(array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
       17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
       34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50,
       51, 52, 53, 54, 55, 56, 57, 58, 59]), array([60, 61, 62, 63]))

Methods

get_params([deep]) Get parameters for this estimator.
set_params(**params) Set the parameters of this estimator.
split(y[, exogenous]) Generate indices to split data into training and test sets
__init__(h=1, step=1, initial=None)[source][source]

Initialize self. See help(type(self)) for accurate signature.

get_params(deep=True)[source]

Get parameters for this estimator.

Parameters:

deep : bool, default=True

If True, will return the parameters for this estimator and contained subobjects that are estimators.

Returns:

params : mapping of string to any

Parameter names mapped to their values.

horizon

The forecast horizon for the cross-validator

set_params(**params)[source]

Set the parameters of this estimator.

The method works on simple estimators as well as on nested objects (such as pipelines). The latter have parameters of the form <component>__<parameter> so that it’s possible to update each component of a nested object.

Parameters:

**params : dict

Estimator parameters.

Returns:

self : object

Estimator instance.

split(y, exogenous=None)[source]

Generate indices to split data into training and test sets

Parameters:

y : array-like or iterable, shape=(n_samples,)

The time-series array.

exogenous : array-like, shape=[n_obs, n_vars], optional (default=None)

An optional 2-d array of exogenous variables.

Yields:

train : np.ndarray

The training set indices for the split

test : np.ndarray

The test set indices for the split