def predict(self, X):
"""Predict using provided Sklearn compatible regressor.
If `optimize_for_horizon` is set to True, then new model is created for
each new horizon and fitted independently
(i.e. len(X)=5 --> horizon=5 --> 5 models will be fitted).
The final prediction is then combination of single point forecast of individual models
for different horizons.
Parameters
----------
X : pandas.DataFrame
Input features.
Returns
-------
pandas.DataFrame
Prediction is stored in column with name being the `name` of the wrapper.
"""
if len(X) > len(self._X) + 3:
raise InsufficientDataLengthError(
f"`X` must have at least {len(self._X) + 3} observations. Please provide valid data."
)
if self.optimize_for_horizon:
preds = pd.concat(
[self._predict(X.iloc[:index, :]).tail(1) for index in range(1, X.shape[0] + 1)]
)
else:
preds = self._predict(X)
preds.index = X.index
return self._clip_predictions(preds)
def _check_X_y(self, X, y=None):
if not isinstance(X, pd.DataFrame):
raise TypeError("`X` must be a pandas dataframe.")
if len(X) < 3:
raise InsufficientDataLengthError(
"`X` must have at least 3 observations. "
"Please provide valid data.")
if not pd.api.types.is_datetime64_dtype(X.index):
raise ValueError(
f"`X` must contain index of type datetime. Your index is {X.index}"
)
if y is not None:
if not isinstance(y, (pd.Series, np.ndarray)):
raise TypeError(
f"`y` must be either pandas series or numpy ndarray. "
f"You provided `{type(y)}`")
if len(X) != len(y):
raise ValueError(f"`X` and `y` must have same length. "
f"`len(X)={len(X)}` and `len(y)={len(y)}`")
if y.ndim != 1:
raise ValueError(f"`y` must have 1 dimension. "
f"You provided y with ndim={y.ndim}")
return func(self, X, y)
def _transform_data_to_tsmodel_input_format(self, X, y=None, horizon=None):
"""Trasnform data into Sklearn API required form and shift them.
Shift is done in autoregressive format with `lags` columns based on prediction horizon which
is derived from length of provided input data for `predict` call.
Parameters
----------
X : pandas.DataFrame
Input features.
y : array_like, (1d)
Target vector
horizon: int
Number of steps used to shift the data
Returns
-------
X, y
X - pandas.DataFrame
y - numpy.ndarray
"""
if y is not None:
if self.lags + horizon > len(self._y):
raise InsufficientDataLengthError(
f"Sum of model lags ({self.lags}) and forecasting horizon ({horizon} "
f"cannot be bigger than length of y ({len(y)}).")
y = self._y[self.lags + horizon - 1:]
X = self._add_lag_features(X, self._y, horizon)
if X.filter(like="_holiday_").shape[1] > 0:
X = self._adjust_holidays(X)
X = X.astype(float)
return X, y