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 crossvalidator See also
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 crossvalidator

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 : arraylike or iterable, shape=(n_samples,)
The timeseries array.
exogenous : arraylike, shape=[n_obs, n_vars], optional (default=None)
An optional 2d 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
