ForecasterAutoregMultiSeries¶
ForecasterAutoregMultiSeries (ForecasterBase)
¶
This class turns any regressor compatible with the scikit-learn API into a
recursive autoregressive (multi-step) forecaster for multiple series.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
regressor |
object |
An instance of a regressor or pipeline compatible with the scikit-learn API. |
required |
lags |
Union[int, numpy.ndarray, list] |
Lags used as predictors. Index starts at 1, so lag 1 is equal to t-1.
|
required |
transformer_series |
Union[object, dict] |
An instance of a transformer (preprocessor) compatible with the scikit-learn
preprocessing API with methods: fit, transform, fit_transform and inverse_transform.
If a single transformer is passed, it is cloned and applied to all series. If a
dict, a different transformer can be used for each series. Transformation is
applied to each |
None |
transformer_exog |
Optional[object] |
An instance of a transformer (preprocessor) compatible with the scikit-learn
preprocessing API. The transformation is applied to |
None |
weight_func |
Union[<built-in function callable>, dict] |
Function that defines the individual weights for each sample based on the
index. For example, a function that assigns a lower weight to certain dates.
If dict {'series_column_name' : callable} a different function can be
used for each series, a weight of 1 is given to all series not present
in |
None |
series_weights |
Optional[dict] |
Weights associated with each series {'series_column_name' : float}. It is only
applied if the |
None |
Attributes:
| Name | Type | Description |
|---|---|---|
regressor |
regressor or pipeline compatible with the scikit-learn API |
An instance of a regressor or pipeline compatible with the scikit-learn API. |
lags |
numpy ndarray |
Lags used as predictors. |
transformer_series |
transformer (preprocessor) or dict of transformers, default `None` |
An instance of a transformer (preprocessor) compatible with the scikit-learn
preprocessing API with methods: fit, transform, fit_transform and inverse_transform.
If a single transformer is passed, it is cloned and applied to all series. If a
dict, a different transformer can be used for each series. Transformation is
applied to each |
transformer_series_ |
dict |
Dictionary with the transformer for each series. It is created cloning the objects
in |
transformer_exog |
transformer (preprocessor), default `None` |
An instance of a transformer (preprocessor) compatible with the scikit-learn
preprocessing API. The transformation is applied to |
weight_func |
callable, dict, default `None` |
Function that defines the individual weights of each sample based on the
index. For example, a function that assigns a lower weight to certain dates.
If dict {'series_column_name': callable} a different function can be
used for each series, a weight of 1 is given to all series not present
in |
weight_func_ |
dict |
Dictionary with the |
source_code_weight_func |
str, dict |
Source code of the custom function(s) used to create weights. New in version 0.6.0 |
series_weights |
dict, default `None` |
Weights associated with each series {'series_column_name': float}. It is only
applied if the |
series_weights_ |
dict |
Weights associated with each series.It is created as a clone of |
max_lag |
int |
Maximum value of lag included in |
window_size |
int |
Size of the window needed to create the predictors. It is equal to
|
last_window |
pandas Series |
Last window the forecaster has seen during training. It stores the
values needed to predict the next |
index_type |
type |
Type of index of the input used in training. |
index_freq |
str |
Frequency of Index of the input used in training. |
index_values |
pandas Index |
Values of Index of the input used in training. |
training_range |
pandas Index |
First and last values of index of the data used during training. |
included_exog |
bool |
If the forecaster has been trained using exogenous variable/s. |
exog_type |
type |
Type of exogenous variable/s used in training. |
exog_col_names |
list |
Names of columns of |
series_col_names |
list |
Names of the series (levels) used during training. |
X_train_col_names |
list |
Names of columns of the matrix created internally for training. |
in_sample_residuals |
dict |
Residuals of the model when predicting training data. Only stored up to
1000 values in the form |
out_sample_residuals |
dict |
Residuals of the model when predicting non-training data. Only stored
up to 1000 values in the form |
fitted |
Bool |
Tag to identify if the regressor has been fitted (trained). |
creation_date |
str |
Date of creation. |
fit_date |
str |
Date of last fit. |
skforcast_version |
str |
Version of skforecast library used to create the forecaster. |
python_version |
str |
Version of python used to create the forecaster. |
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
class ForecasterAutoregMultiSeries(ForecasterBase):
"""
This class turns any regressor compatible with the scikit-learn API into a
recursive autoregressive (multi-step) forecaster for multiple series.
Parameters
----------
regressor : regressor or pipeline compatible with the scikit-learn API
An instance of a regressor or pipeline compatible with the scikit-learn API.
lags : int, list, numpy ndarray, range
Lags used as predictors. Index starts at 1, so lag 1 is equal to t-1.
`int`: include lags from 1 to `lags` (included).
`list`, `numpy ndarray` or `range`: include only lags present in `lags`,
all elements must be int.
transformer_series : transformer (preprocessor) or dict of transformers, default `None`
An instance of a transformer (preprocessor) compatible with the scikit-learn
preprocessing API with methods: fit, transform, fit_transform and inverse_transform.
If a single transformer is passed, it is cloned and applied to all series. If a
dict, a different transformer can be used for each series. Transformation is
applied to each `series` before training the forecaster.
ColumnTransformers are not allowed since they do not have inverse_transform method.
transformer_exog : transformer, default `None`
An instance of a transformer (preprocessor) compatible with the scikit-learn
preprocessing API. The transformation is applied to `exog` before training the
forecaster. `inverse_transform` is not available when using ColumnTransformers.
weight_func : callable, dict, default `None`
Function that defines the individual weights for each sample based on the
index. For example, a function that assigns a lower weight to certain dates.
If dict {'series_column_name' : callable} a different function can be
used for each series, a weight of 1 is given to all series not present
in `weight_func`. Ignored if `regressor` does not have the argument
`sample_weight` in its `fit` method. See Notes section for more details
on the use of the weights.
**New in version 0.6.0**
series_weights : dict, default `None`
Weights associated with each series {'series_column_name' : float}. It is only
applied if the `regressor` used accepts `sample_weight` in its `fit` method.
If `series_weights` is provided, a weight of 1 is given to all series not present
in `series_weights`. If `None`, all levels have the same weight. See Notes section
for more details on the use of the weights.
**New in version 0.6.0**
Attributes
----------
regressor : regressor or pipeline compatible with the scikit-learn API
An instance of a regressor or pipeline compatible with the scikit-learn API.
lags : numpy ndarray
Lags used as predictors.
transformer_series : transformer (preprocessor) or dict of transformers, default `None`
An instance of a transformer (preprocessor) compatible with the scikit-learn
preprocessing API with methods: fit, transform, fit_transform and inverse_transform.
If a single transformer is passed, it is cloned and applied to all series. If a
dict, a different transformer can be used for each series. Transformation is
applied to each `series` before training the forecaster.
ColumnTransformers are not allowed since they do not have inverse_transform method.
transformer_series_ : dict
Dictionary with the transformer for each series. It is created cloning the objects
in `transformer_series` and is used internally to avoid overwriting.
transformer_exog : transformer (preprocessor), default `None`
An instance of a transformer (preprocessor) compatible with the scikit-learn
preprocessing API. The transformation is applied to `exog` before training the
forecaster. `inverse_transform` is not available when using ColumnTransformers.
weight_func : callable, dict, default `None`
Function that defines the individual weights of each sample based on the
index. For example, a function that assigns a lower weight to certain dates.
If dict {'series_column_name': callable} a different function can be
used for each series, a weight of 1 is given to all series not present
in `weight_func`. Ignored if `regressor` does not have the argument
`sample_weight` in its `fit` method. See Notes section for more details
on the use of the weights.
**New in version 0.6.0**
weight_func_ : dict
Dictionary with the `weight_func` for each series. It is created cloning the objects
in `weight_func` and is used internally to avoid overwriting.
**New in version 0.6.0**
source_code_weight_func : str, dict
Source code of the custom function(s) used to create weights.
**New in version 0.6.0**
series_weights : dict, default `None`
Weights associated with each series {'series_column_name': float}. It is only
applied if the `regressor` used accepts `sample_weight` in its `fit` method.
If `series_weights` is provided, a weight of 1 is given to all series not present
in `series_weights`. If `None`, all levels have the same weight. See Notes section
for more details on the use of the weights.
**New in version 0.6.0**
series_weights_ : dict
Weights associated with each series.It is created as a clone of `series_weights`
and is used internally to avoid overwriting.
**New in version 0.6.0**
max_lag : int
Maximum value of lag included in `lags`.
window_size : int
Size of the window needed to create the predictors. It is equal to
`max_lag`.
last_window : pandas Series
Last window the forecaster has seen during training. It stores the
values needed to predict the next `step` right after the training data.
index_type : type
Type of index of the input used in training.
index_freq : str
Frequency of Index of the input used in training.
index_values : pandas Index
Values of Index of the input used in training.
training_range: pandas Index
First and last values of index of the data used during training.
included_exog : bool
If the forecaster has been trained using exogenous variable/s.
exog_type : type
Type of exogenous variable/s used in training.
exog_col_names : list
Names of columns of `exog` if `exog` used in training was a pandas
DataFrame.
series_col_names : list
Names of the series (levels) used during training.
X_train_col_names : list
Names of columns of the matrix created internally for training.
in_sample_residuals : dict
Residuals of the model when predicting training data. Only stored up to
1000 values in the form `{level: residuals}`.
out_sample_residuals : dict
Residuals of the model when predicting non-training data. Only stored
up to 1000 values in the form `{level: residuals}`. Use
`set_out_sample_residuals` to set values.
fitted : Bool
Tag to identify if the regressor has been fitted (trained).
creation_date : str
Date of creation.
fit_date : str
Date of last fit.
skforcast_version : str
Version of skforecast library used to create the forecaster.
python_version : str
Version of python used to create the forecaster.
Notes
-----
The weights are used to control the influence that each observation has on the
training of the model. `ForecasterAutoregMultiseries` accepts two types of weights:
+ series_weights : controls the relative importance of each series. If a series has
twice as much weight as the others, the observations of that series influence the
training twice as much. The higher the weight of a series relative to the others,
the more the model will focus on trying to learn that series.
+ weight_func : controls the relative importance of each observation according to its
index value. For example, a function that assigns a lower weight to certain dates.
If the two types of weights are indicated, they are multiplied to create the final
weights. The resulting `sample_weight` cannot have negative values.
"""
def __init__(
self,
regressor: object,
lags: Union[int, np.ndarray, list],
transformer_series: Optional[Union[object, dict]]=None,
transformer_exog: Optional[object]=None,
weight_func: Optional[Union[callable, dict]]=None,
series_weights: Optional[dict]=None
) -> None:
self.regressor = regressor
self.transformer_series = transformer_series
self.transformer_series_ = None
self.transformer_exog = transformer_exog
self.weight_func = weight_func
self.weight_func_ = None
self.source_code_weight_func = None
self.series_weights = series_weights
self.series_weights_ = None
self.index_type = None
self.index_freq = None
self.index_values = None
self.training_range = None
self.last_window = None
self.included_exog = False
self.exog_type = None
self.exog_col_names = None
self.series_col_names = None
self.X_train_col_names = None
self.in_sample_residuals = None
self.out_sample_residuals = None
self.fitted = False
self.creation_date = pd.Timestamp.today().strftime('%Y-%m-%d %H:%M:%S')
self.fit_date = None
self.skforcast_version = skforecast.__version__
self.python_version = sys.version.split(" ")[0]
self.lags = initialize_lags(type(self).__name__, lags)
self.max_lag = max(self.lags)
self.window_size = self.max_lag
self.weight_func, self.source_code_weight_func, self.series_weights = initialize_weights(
forecaster_type = type(self).__name__,
regressor = regressor,
weight_func = weight_func,
series_weights = series_weights
)
def __repr__(
self
) -> str:
"""
Information displayed when a ForecasterAutoregMultiSeries object is printed.
"""
if isinstance(self.regressor, sklearn.pipeline.Pipeline):
name_pipe_steps = tuple(name + "__" for name in self.regressor.named_steps.keys())
params = {key : value for key, value in self.regressor.get_params().items() \
if key.startswith(name_pipe_steps)}
else:
params = self.regressor.get_params()
info = (
f"{'=' * len(str(type(self)).split('.')[1])} \n"
f"{str(type(self)).split('.')[1]} \n"
f"{'=' * len(str(type(self)).split('.')[1])} \n"
f"Regressor: {self.regressor} \n"
f"Lags: {self.lags} \n"
f"Transformer for series: {self.transformer_series} \n"
f"Transformer for exog: {self.transformer_exog} \n"
f"Window size: {self.window_size} \n"
f"Series levels (names): {self.series_col_names} \n"
f"Series weights: {self.series_weights} \n"
f"Weight function included: {True if self.weight_func is not None else False} \n"
f"Exogenous included: {self.included_exog} \n"
f"Type of exogenous variable: {self.exog_type} \n"
f"Exogenous variables names: {self.exog_col_names} \n"
f"Training range: {self.training_range.to_list() if self.fitted else None} \n"
f"Training index type: {str(self.index_type).split('.')[-1][:-2] if self.fitted else None} \n"
f"Training index frequency: {self.index_freq if self.fitted else None} \n"
f"Regressor parameters: {params} \n"
f"Creation date: {self.creation_date} \n"
f"Last fit date: {self.fit_date} \n"
f"Skforecast version: {self.skforcast_version} \n"
f"Python version: {self.python_version} \n"
)
return info
def _create_lags(
self,
y: np.ndarray
) -> Tuple[np.ndarray, np.ndarray]:
"""
Transforms a 1d array into a 2d array (X) and a 1d array (y). Each row
in X is associated with a value of y and it represents the lags that
precede it.
Notice that, the returned matrix X_data, contains the lag 1 in the first
column, the lag 2 in the second column and so on.
Parameters
----------
y : 1d numpy ndarray
Training time series.
Returns
-------
X_data : 2d numpy ndarray, shape (samples - max(self.lags), len(self.lags))
2d numpy array with the lagged values (predictors).
y_data : 1d numpy ndarray, shape (samples - max(self.lags),)
Values of the time series related to each row of `X_data`.
"""
n_splits = len(y) - self.max_lag
if n_splits <= 0:
raise ValueError(
f'The maximum lag ({self.max_lag}) must be less than the length '
f'of the series ({len(y)}).'
)
X_data = np.full(shape=(n_splits, len(self.lags)), fill_value=np.nan, dtype=float)
for i, lag in enumerate(self.lags):
X_data[:, i] = y[self.max_lag - lag: -lag]
y_data = y[self.max_lag:]
return X_data, y_data
def create_train_X_y(
self,
series: pd.DataFrame,
exog: Optional[Union[pd.Series, pd.DataFrame]]=None
) -> Tuple[pd.DataFrame, pd.Series, pd.Index, pd.Index]:
"""
Create training matrices from multiple time series and exogenous
variables.
Parameters
----------
series : pandas DataFrame
Training time series.
exog : pandas Series, pandas DataFrame, default `None`
Exogenous variable/s included as predictor/s. Must have the same
number of observations as `series` and their indexes must be aligned.
Returns
-------
X_train : pandas DataFrame
Pandas DataFrame with the training values (predictors).
y_train : pandas Series, shape (len(series) - self.max_lag, )
Values (target) of the time series related to each row of `X_train`.
y_index : pandas Index
Index of `series`.
y_train_index: pandas Index
Index of `y_train`.
"""
if not isinstance(series, pd.DataFrame):
raise TypeError(f'`series` must be a pandas DataFrame. Got {type(series)}.')
series_col_names = list(series.columns)
if self.transformer_series is None:
self.transformer_series_ = {serie: None for serie in series_col_names}
elif not isinstance(self.transformer_series, dict):
self.transformer_series_ = {serie: clone(self.transformer_series)
for serie in series_col_names}
else:
self.transformer_series_ = {serie: None for serie in series_col_names}
# Only elements already present in transformer_series_ are updated
self.transformer_series_.update(
(k, v) for k, v in deepcopy(self.transformer_series).items() if k in self.transformer_series_
)
series_not_in_transformer_series = set(series.columns) - set(self.transformer_series.keys())
if series_not_in_transformer_series:
warnings.warn(
f"{series_not_in_transformer_series} not present in `transformer_series`."
f" No transformation is applied to these series."
)
X_levels = []
X_train_col_names = [f"lag_{lag}" for lag in self.lags]
for i, serie in enumerate(series.columns):
y = series[serie]
check_y(y=y)
y = transform_series(
series = y,
transformer = self.transformer_series_[serie],
fit = True,
inverse_transform = False
)
y_values, y_index = preprocess_y(y=y)
X_train_values, y_train_values = self._create_lags(y=y_values)
if i == 0:
X_train = X_train_values
y_train = y_train_values
else:
X_train = np.vstack((X_train, X_train_values))
y_train = np.append(y_train, y_train_values)
X_level = [serie]*len(X_train_values)
X_levels.extend(X_level)
if exog is not None:
if len(exog) != len(series):
raise ValueError(
f'`exog` must have same number of samples as `series`. '
f'length `exog`: ({len(exog)}), length `series`: ({len(series)})'
)
check_exog(exog=exog)
if isinstance(exog, pd.Series):
exog = transform_series(
series = exog,
transformer = self.transformer_exog,
fit = True,
inverse_transform = False
)
else:
exog = transform_dataframe(
df = exog,
transformer = self.transformer_exog,
fit = True,
inverse_transform = False
)
exog_values, exog_index = preprocess_exog(exog=exog)
if not (exog_index[:len(y_index)] == y_index).all():
raise ValueError(
('Different index for `series` and `exog`. They must be equal '
'to ensure the correct alignment of values.')
)
col_names_exog = exog.columns if isinstance(exog, pd.DataFrame) else [exog.name]
X_train_col_names.extend(col_names_exog)
# The first `self.max_lag` positions have to be removed from exog
# since they are not in X_train. Then exog is cloned as many times
# as series.
if exog_values.ndim == 1:
X_train = np.column_stack((
X_train,
np.tile(exog_values[self.max_lag:, ], series.shape[1])
))
else:
X_train = np.column_stack((
X_train,
np.tile(exog_values[self.max_lag:, ], [series.shape[1], 1])
))
X_levels = pd.Series(X_levels)
X_levels = pd.get_dummies(X_levels, dtype=float)
X_train_col_names.extend(X_levels.columns)
X_train = np.column_stack((X_train, X_levels.values))
X_train = pd.DataFrame(
data = X_train,
columns = X_train_col_names
)
y_train = pd.Series(
data = y_train,
name = 'y'
)
y_train_index = pd.Index(
np.tile(
y_index[self.max_lag: ].values,
reps = len(series_col_names)
)
)
self.X_train_col_names = X_train_col_names
return X_train, y_train, y_index, y_train_index
def create_sample_weights(
self,
series: pd.DataFrame,
X_train: pd.DataFrame,
y_train_index: pd.Index,
)-> np.ndarray:
"""
Crate weights for each observation according to the forecaster's attributes
`series_weights` and `weight_func`. The resulting weights are product of both
types of weights.
Parameters
----------
series : pandas DataFrame
Time series used to create `X_train` with the method `create_train_X_y`.
X_train : pandas DataFrame
Dataframe generated with the method `create_train_X_y`, first return.
y_train_index : pandas Index
Index of `y_train` generated with the method `create_train_X_y`, fourth return.
Returns
-------
weights : numpy ndarray
Weights to use in `fit` method.
"""
weights = None
weights_samples = None
weights_series = None
if self.series_weights is not None:
# Series not present in series_weights have a weight of 1 in all their samples.
# Keys in series_weights not present in series are ignored.
series_not_in_series_weights = set(series.columns) - set(self.series_weights.keys())
if series_not_in_series_weights:
warnings.warn(
f"{series_not_in_series_weights} not present in `series_weights`."
f" A weight of 1 is given to all their samples."
)
self.series_weights_ = dict.fromkeys(series.columns, 1.)
self.series_weights_.update((k, v) for k, v in self.series_weights.items() if k in self.series_weights_)
weights_series = [np.repeat(self.series_weights_[serie], sum(X_train[serie]))
for serie in series.columns]
weights_series = np.concatenate(weights_series)
if self.weight_func is not None:
if isinstance(self.weight_func, Callable):
self.weight_func_ = dict.fromkeys(series.columns, self.weight_func)
else:
# Series not present in weight_func have a weight of 1 in all their samples
series_not_in_weight_func = set(series.columns) - set(self.weight_func.keys())
if series_not_in_weight_func:
warnings.warn(
f"{series_not_in_weight_func} not present in `weight_func`."
f" A weight of 1 is given to all their samples."
)
self.weight_func_ = dict.fromkeys(series.columns, lambda index: np.ones_like(index, dtype=float))
self.weight_func_.update((k, v) for k, v in self.weight_func.items() if k in self.weight_func_)
weights_samples = []
for key in self.weight_func_.keys():
index = y_train_index[X_train[X_train[key] == 1.0].index]
weights_samples.append(self.weight_func_[key](index))
weights_samples = np.concatenate(weights_samples)
if weights_series is not None:
weights = weights_series
if weights_samples is not None:
weights = weights * weights_samples
else:
if weights_samples is not None:
weights = weights_samples
if weights is not None:
if np.isnan(weights).any():
raise ValueError(
"The resulting `weights` cannot have NaN values."
)
if np.any(weights < 0):
raise ValueError(
"The resulting `weights` cannot have negative values."
)
if np.sum(weights) == 0:
raise ValueError(
("The resulting `weights` cannot be normalized because "
"the sum of the weights is zero.")
)
return weights
def fit(
self,
series: pd.DataFrame,
exog: Optional[Union[pd.Series, pd.DataFrame]]=None,
store_in_sample_residuals: bool=True
) -> None:
"""
Training Forecaster.
Parameters
----------
series : pandas DataFrame
Training time series.
exog : pandas Series, pandas DataFrame, default `None`
Exogenous variable/s included as predictor/s. Must have the same
number of observations as `series` and their indexes must be aligned so
that series[i] is regressed on exog[i].
store_in_sample_residuals : bool, default `True`
if True, in_sample_residuals are stored.
Returns
-------
None
"""
# Reset values in case the forecaster has already been fitted.
self.index_type = None
self.index_freq = None
self.index_values = None
self.last_window = None
self.included_exog = False
self.exog_type = None
self.exog_col_names = None
self.series_col_names = None
self.X_train_col_names = None
self.in_sample_residuals = None
self.fitted = False
self.training_range = None
self.series_col_names = list(series.columns)
if exog is not None:
self.included_exog = True
self.exog_type = type(exog)
self.exog_col_names = \
exog.columns.to_list() if isinstance(exog, pd.DataFrame) else [exog.name]
if len(set(self.exog_col_names) - set(self.series_col_names)) != len(self.exog_col_names):
raise ValueError(
(f'`exog` cannot contain a column named the same as one of the series'
f' (column names of series).\n'
f' `series` columns : {self.series_col_names}.\n'
f' `exog` columns : {self.exog_col_names}.')
)
X_train, y_train, y_index, y_train_index = self.create_train_X_y(series=series, exog=exog)
sample_weight = self.create_sample_weights(
series = series,
X_train = X_train,
y_train_index = y_train_index,
)
if sample_weight is not None:
if not str(type(self.regressor)) == "<class 'xgboost.sklearn.XGBRegressor'>":
self.regressor.fit(X=X_train, y=y_train, sample_weight=sample_weight)
else:
self.regressor.fit(X=X_train.to_numpy(), y=y_train.to_numpy(), sample_weight=sample_weight)
else:
if not str(type(self.regressor)) == "<class 'xgboost.sklearn.XGBRegressor'>":
self.regressor.fit(X=X_train, y=y_train)
else:
self.regressor.fit(X=X_train.to_numpy(), y=y_train.to_numpy())
self.fitted = True
self.fit_date = pd.Timestamp.today().strftime('%Y-%m-%d %H:%M:%S')
self.training_range = y_index[[0, -1]]
self.index_type = type(y_index)
if isinstance(y_index, pd.DatetimeIndex):
self.index_freq = y_index.freqstr
else:
self.index_freq = y_index.step
self.index_values = y_index
in_sample_residuals = {}
# This is done to save time during fit in functions such as backtesting()
if store_in_sample_residuals:
if not str(type(self.regressor)) == "<class 'xgboost.sklearn.XGBRegressor'>":
residuals = y_train - self.regressor.predict(X_train)
else:
residuals = y_train - self.regressor.predict(X_train.to_numpy())
for serie in series.columns:
in_sample_residuals[serie] = residuals.values[X_train[serie] == 1.]
if len(in_sample_residuals[serie]) > 1000:
# Only up to 1000 residuals are stored
rng = np.random.default_rng(seed=123)
in_sample_residuals[serie] = rng.choice(
a = in_sample_residuals[serie],
size = 1000,
replace = False
)
else:
for serie in series.columns:
in_sample_residuals[serie] = np.array([None])
self.in_sample_residuals = in_sample_residuals
# The last time window of training data is stored so that lags needed as
# predictors in the first iteration of `predict()` can be calculated.
self.last_window = series.iloc[-self.max_lag:, ].copy()
def _recursive_predict(
self,
steps: int,
level: str,
last_window: np.ndarray,
exog: Optional[np.ndarray]=None
) -> np.ndarray:
"""
Predict n steps ahead. It is an iterative process in which, each prediction,
is used as a predictor for the next step.
Parameters
----------
steps : int
Number of future steps predicted.
level : str
Time series to be predicted.
last_window : numpy ndarray
Values of the series used to create the predictors (lags) need in the
first iteration of prediction (t + 1).
exog : numpy ndarray, default `None`
Exogenous variable/s included as predictor/s.
Returns
-------
predictions : numpy ndarray
Predicted values.
"""
predictions = np.full(shape=steps, fill_value=np.nan)
for i in range(steps):
X = last_window[-self.lags].reshape(1, -1)
if exog is not None:
X = np.column_stack((X, exog[i, ].reshape(1, -1)))
levels_dummies = np.zeros(shape=(1, len(self.series_col_names)), dtype=float)
levels_dummies[0][self.series_col_names.index(level)] = 1.
X = np.column_stack((X, levels_dummies.reshape(1, -1)))
with warnings.catch_warnings():
# Suppress scikit-learn warning: "X does not have valid feature names,
# but NoOpTransformer was fitted with feature names".
warnings.simplefilter("ignore")
prediction = self.regressor.predict(X)
predictions[i] = prediction.ravel()[0]
# Update `last_window` values. The first position is discarded and
# the new prediction is added at the end.
last_window = np.append(last_window[1:], prediction)
return predictions
def predict(
self,
steps: int,
levels: Optional[Union[str, list]]=None,
last_window: Optional[pd.DataFrame]=None,
exog: Optional[Union[pd.Series, pd.DataFrame]]=None
) -> pd.DataFrame:
"""
Predict n steps ahead. It is an recursive process in which, each prediction,
is used as a predictor for the next step.
Parameters
----------
steps : int
Number of future steps predicted.
levels : str, list, default `None`
Time series to be predicted. If `None` all levels will be predicted.
**New in version 0.6.0**
last_window : pandas DataFrame, default `None`
Values of the series used to create the predictors (lags) need in the
first iteration of prediction (t + 1).
If `last_window = None`, the values stored in `self.last_window` are
used to calculate the initial predictors, and the predictions start
right after training data.
exog : pandas Series, pandas DataFrame, default `None`
Exogenous variable/s included as predictor/s.
Returns
-------
predictions : pandas DataFrame
Predicted values, one column for each level.
"""
if levels is None:
levels = self.series_col_names
elif isinstance(levels, str):
levels = [levels]
check_predict_input(
forecaster_type = type(self).__name__,
steps = steps,
fitted = self.fitted,
included_exog = self.included_exog,
index_type = self.index_type,
index_freq = self.index_freq,
window_size = self.window_size,
last_window = last_window,
exog = exog,
exog_type = self.exog_type,
exog_col_names = self.exog_col_names,
interval = None,
max_steps = None,
levels = levels,
series_col_names = self.series_col_names
)
if exog is not None:
if isinstance(exog, pd.DataFrame):
exog = transform_dataframe(
df = exog,
transformer = self.transformer_exog,
fit = False,
inverse_transform = False
)
else:
exog = transform_series(
series = exog,
transformer = self.transformer_exog,
fit = False,
inverse_transform = False
)
exog_values, _ = preprocess_exog(
exog = exog.iloc[:steps, ]
)
else:
exog_values = None
predictions = []
for level in levels:
if last_window is None:
last_window = self.last_window.copy()
last_window_level = transform_series(
series = last_window[level],
transformer = self.transformer_series_[level],
fit = False,
inverse_transform = False
)
last_window_values, last_window_index = preprocess_last_window(
last_window = last_window_level
)
preds_level = self._recursive_predict(
steps = steps,
level = level,
last_window = copy(last_window_values),
exog = copy(exog_values)
)
preds_level = pd.Series(
data = preds_level,
index = expand_index(
index = last_window_index,
steps = steps
),
name = level
)
preds_level = transform_series(
series = preds_level,
transformer = self.transformer_series_[level],
fit = False,
inverse_transform = True
)
predictions.append(preds_level)
predictions = pd.concat(predictions, axis=1)
return predictions
def _estimate_boot_interval(
self,
steps: int,
level: str,
last_window: Optional[np.ndarray]=None,
exog: Optional[np.ndarray]=None,
interval: list=[5, 95],
n_boot: int=500,
random_state: int=123,
in_sample_residuals: bool=True
) -> np.ndarray:
"""
Iterative process in which, each prediction, is used as a predictor
for the next step and bootstrapping is used to estimate prediction
intervals. This method only returns prediction intervals.
See predict_intervals() to calculate both, predictions and intervals.
Parameters
----------
steps : int
Number of future steps predicted.
level : str
Time series to be predicted.
last_window : 1d numpy ndarray shape (, max_lag), default `None`
Values of the series used to create the predictors (lags) needed in the
first iteration of prediction (t + 1).
If `last_window = `None`, the values stored in` self.last_window` are
used to calculate the initial predictors, and the predictions start
right after training data.
exog : numpy ndarray, default `None`
Exogenous variable/s included as predictor/s.
n_boot : int, default `500`
Number of bootstrapping iterations used to estimate prediction
intervals.
random_state : int
Sets a seed to the random generator, so that boot intervals are always
deterministic.
interval : list, default `[5, 95]`
Confidence of the prediction interval estimated. Sequence of
percentiles to compute, which must be between 0 and 100 inclusive.
For example, interval of 95% should be as `interval = [2.5, 97.5]`.
in_sample_residuals : bool, default `True`
If `True`, residuals from the training data are used as proxy of
prediction error to create prediction intervals. If `False`, out of
sample residuals are used. In the latter case, the user should have
calculated and stored the residuals within the forecaster (see
`set_out_sample_residuals()`).
Returns
-------
prediction_interval : numpy ndarray, shape (steps, 2)
Interval estimated for each prediction by bootstrapping:
lower_bound: lower bound of the interval.
upper_bound: upper bound interval of the interval.
Notes
-----
More information about prediction intervals in forecasting:
https://otexts.com/fpp2/prediction-intervals.html
Forecasting: Principles and Practice (2nd ed) Rob J Hyndman and
George Athanasopoulos.
"""
if last_window is None:
last_window = self.last_window[level]
last_window = last_window.values
boot_predictions = np.full(
shape = (steps, n_boot),
fill_value = np.nan,
dtype = float
)
rng = np.random.default_rng(seed=random_state)
seeds = rng.integers(low=0, high=10000, size=n_boot)
for i in range(n_boot):
# In each bootstraping iteration the initial last_window and exog
# need to be restored.
last_window_boot = last_window.copy()
if exog is not None:
exog_boot = exog.copy()
else:
exog_boot = None
if in_sample_residuals:
residuals = self.in_sample_residuals[level]
else:
residuals = self.out_sample_residuals[level]
rng = np.random.default_rng(seed=seeds[i])
sample_residuals = rng.choice(
a = residuals,
size = steps,
replace = True
)
for step in range(steps):
prediction = self._recursive_predict(
steps = 1,
level = level,
last_window = last_window_boot,
exog = exog_boot
)
prediction_with_residual = prediction + sample_residuals[step]
boot_predictions[step, i] = prediction_with_residual
last_window_boot = np.append(
last_window_boot[1:],
prediction_with_residual
)
if exog is not None:
exog_boot = exog_boot[1:]
prediction_interval = np.percentile(boot_predictions, q=interval, axis=1)
prediction_interval = prediction_interval.transpose()
return prediction_interval
def predict_interval(
self,
steps: int,
levels: Optional[Union[str, list]]=None,
last_window: Optional[pd.DataFrame]=None,
exog: Optional[Union[pd.Series, pd.DataFrame]]=None,
interval: list=[5, 95],
n_boot: int=500,
random_state: int=123,
in_sample_residuals: bool=True
) -> pd.DataFrame:
"""
Iterative process in which, each prediction, is used as a predictor
for the next step and bootstrapping is used to estimate prediction
intervals. Both, predictions and intervals, are returned.
Parameters
----------
steps : int
Number of future steps predicted.
levels : str, list, default `None`
Time series to be predicted. If `None` all levels will be predicted.
**New in version 0.6.0**
last_window : pandas DataFrame, default `None`
Values of the series used to create the predictors (lags) needed in the
first iteration of prediction (t + 1).
If `last_window = None`, the values stored in` self.last_window` are
used to calculate the initial predictors, and the predictions start
right after training data.
exog : pandas Series, pandas DataFrame, default `None`
Exogenous variable/s included as predictor/s.
interval : list, default `[5, 95]`
Confidence of the prediction interval estimated. Sequence of
percentiles to compute, which must be between 0 and 100 inclusive.
For example, interval of 95% should be as `interval = [2.5, 97.5]`.
n_boot : int, default `500`
Number of bootstrapping iterations used to estimate prediction
intervals.
random_state : int, default 123
Sets a seed to the random generator, so that boot intervals are always
deterministic.
in_sample_residuals : bool, default `True`
If `True`, residuals from the training data are used as proxy of
prediction error to create prediction intervals. If `False`, out of
sample residuals are used. In the latter case, the user should have
calculated and stored the residuals within the forecaster (see
`set_out_sample_residuals()`).
Returns
-------
predictions : pandas DataFrame
Values predicted by the forecaster and their estimated interval.
level: predictions.
level_lower_bound: lower bound of the interval.
level_upper_bound: upper bound interval of the interval.
Notes
-----
More information about prediction intervals in forecasting:
https://otexts.com/fpp2/prediction-intervals.html
Forecasting: Principles and Practice (2nd ed) Rob J Hyndman and
George Athanasopoulos.
"""
if levels is None:
levels = self.series_col_names
elif isinstance(levels, str):
levels = [levels]
for level in levels:
if in_sample_residuals and (self.in_sample_residuals[level] == None).any():
raise ValueError(
(f"`forecaster.in_sample_residuals['{level}']` contains `None` values. "
"Try using `fit` method with `in_sample_residuals=True` or set in "
"`predict_interval` method `in_sample_residuals=False` and use "
"`out_sample_residuals` (see `set_out_sample_residuals()`).")
)
if not in_sample_residuals and self.out_sample_residuals is None:
raise ValueError(
('`forecaster.out_sample_residuals` is `None`. Use '
'`in_sample_residuals=True` or method `set_out_sample_residuals()` '
'before `predict_interval()`.')
)
if not in_sample_residuals and len(set(levels) - set(self.out_sample_residuals.keys())) != 0:
raise ValueError(
(f'Not `forecaster.out_sample_residuals` for levels: {set(levels) - set(self.out_sample_residuals.keys())}. '
f'Use method `set_out_sample_residuals()`.')
)
check_predict_input(
forecaster_type = type(self).__name__,
steps = steps,
fitted = self.fitted,
included_exog = self.included_exog,
index_type = self.index_type,
index_freq = self.index_freq,
window_size = self.window_size,
last_window = last_window,
exog = exog,
exog_type = self.exog_type,
exog_col_names = self.exog_col_names,
interval = interval,
max_steps = None,
levels = levels,
series_col_names = self.series_col_names
)
if exog is not None:
if isinstance(exog, pd.DataFrame):
exog = transform_dataframe(
df = exog,
transformer = self.transformer_exog,
fit = False,
inverse_transform = False
)
else:
exog = transform_series(
series = exog,
transformer = self.transformer_exog,
fit = False,
inverse_transform = False
)
exog_values, _ = preprocess_exog(
exog = exog.iloc[:steps, ]
)
else:
exog_values = None
predictions = []
for level in levels:
if last_window is None:
last_window = self.last_window.copy()
last_window_level = transform_series(
series = last_window[level],
transformer = self.transformer_series_[level],
fit = False,
inverse_transform = False
)
last_window_values, last_window_index = preprocess_last_window(
last_window = last_window_level
)
# Since during predict() `last_window_values` and `exog_values` are modified,
# the originals are stored to be used later.
last_window_values_original = last_window_values.copy()
if exog is not None:
exog_values_original = exog_values.copy()
else:
exog_values_original = None
preds_level = self._recursive_predict(
steps = steps,
level = level,
last_window = last_window_values,
exog = exog_values
)
preds_level_interval = self._estimate_boot_interval(
steps = steps,
level = level,
last_window = copy(last_window_values_original),
exog = copy(exog_values_original),
interval = interval,
n_boot = n_boot,
random_state = random_state,
in_sample_residuals = in_sample_residuals
)
preds_level = np.column_stack((preds_level, preds_level_interval))
preds_level = pd.DataFrame(
data = preds_level,
index = expand_index(
index = last_window_index,
steps = steps
),
columns = [level, f'{level}_lower_bound', f'{level}_upper_bound']
)
if self.transformer_series_[level]:
for col in preds_level.columns:
preds_level[col] = self.transformer_series_[level].inverse_transform(preds_level[[col]])
predictions.append(preds_level)
predictions = pd.concat(predictions, axis=1)
return predictions
def set_params(
self,
**params: dict
) -> None:
"""
Set new values to the parameters of the scikit learn model stored in the
forecaster.
Parameters
----------
params : dict
Parameters values.
Returns
-------
self
"""
self.regressor = clone(self.regressor)
self.regressor.set_params(**params)
def set_lags(
self,
lags: Union[int, list, np.ndarray, range]
) -> None:
"""
Set new value to the attribute `lags`.
Attributes `max_lag` and `window_size` are also updated.
Parameters
----------
lags : int, list, 1D np.array, range
Lags used as predictors. Index starts at 1, so lag 1 is equal to t-1.
`int`: include lags from 1 to `lags`.
`list` or `np.array`: include only lags present in `lags`.
Returns
-------
None
"""
self.lags = initialize_lags(type(self).__name__, lags)
self.max_lag = max(self.lags)
self.window_size = max(self.lags)
def set_out_sample_residuals(
self,
residuals: pd.DataFrame,
append: bool=True,
transform: bool=True,
random_state: int=123
)-> None:
"""
Set new values to the attribute `out_sample_residuals`. Out of sample
residuals are meant to be calculated using observations that did not
participate in the training process.
Parameters
----------
residuals : pandas DataFrame
Values of residuals. If len(residuals) > 1000, only a random sample
of 1000 values are stored. Columns must be the same as `levels`.
append : bool, default `True`
If `True`, new residuals are added to the once already stored in the
attribute `out_sample_residuals`. Once the limit of 1000 values is
reached, no more values are appended. If False, `out_sample_residuals`
is overwritten with the new residuals.
transform : bool, default `True`
If `True`, new residuals are transformed using self.transformer_series.
random_state : int, default `123`
Sets a seed to the random sampling for reproducible output.
Returns
-------
self
"""
if not isinstance(residuals, pd.DataFrame):
raise TypeError(
f"`residuals` argument must be a pandas DataFrame. Got {type(residuals)}."
)
if not self.fitted:
raise sklearn.exceptions.NotFittedError(
("This forecaster is not fitted yet. Call `fit` with appropriate "
"arguments before using `set_out_sample_residuals()`.")
)
out_sample_residuals = {}
for level in residuals.columns:
if not level in self.series_col_names:
continue
else:
residuals_level = residuals[level]
if not transform and self.transformer_series_[level] is not None:
warnings.warn(
('Argument `transform` is set to `False` but forecaster was trained '
f'using a transformer {self.transformer_series_[level]} for level {level}. '
'Ensure that the new residuals are already transformed or set `transform=True`.')
)
if transform and self.transformer_series_ and self.transformer_series_[level]:
warnings.warn(
('Residuals will be transformed using the same transformer used '
f'when training the forecaster for level {level} : ({self.transformer_series_[level]}). '
'Ensure that the new residuals are on the same scale as the '
'original time series. ')
)
residuals_level = transform_series(
series = residuals_level,
transformer = self.transformer_series_[level],
fit = False,
inverse_transform = False
)
if len(residuals_level) > 1000:
rng = np.random.default_rng(seed=random_state)
residuals_level = rng.choice(a=residuals_level, size=1000, replace=False)
if append and self.out_sample_residuals is not None:
if not level in self.out_sample_residuals.keys():
raise ValueError(
f'{level} does not exists in `forecaster.out_sample_residuals` keys: {list(self.out_sample_residuals.keys())}'
)
free_space = max(0, 1000 - len(self.out_sample_residuals[level]))
if len(residuals_level) < free_space:
residuals_level = np.hstack((
self.out_sample_residuals[level],
residuals_level
))
else:
residuals_level = np.hstack((
self.out_sample_residuals[level],
residuals_level[:free_space]
))
out_sample_residuals[level] = np.array(residuals_level)
self.out_sample_residuals = out_sample_residuals
def get_feature_importance(
self
) -> pd.DataFrame:
"""
Return feature importance of the regressor stored in the
forecaster. Only valid when regressor stores internally the feature
importance in the attribute `feature_importances_` or `coef_`.
Parameters
----------
self
Returns
-------
feature_importance : pandas DataFrame
Feature importance associated with each predictor.
"""
if self.fitted == False:
raise sklearn.exceptions.NotFittedError(
("This forecaster is not fitted yet. Call `fit` with appropriate "
"arguments before using `get_feature_importance()`.")
)
if isinstance(self.regressor, sklearn.pipeline.Pipeline):
estimator = self.regressor[-1]
else:
estimator = self.regressor
try:
feature_importance = pd.DataFrame({
'feature': self.X_train_col_names,
'importance' : estimator.feature_importances_
})
except:
try:
feature_importance = pd.DataFrame({
'feature': self.X_train_col_names,
'importance' : estimator.coef_
})
except:
warnings.warn(
f"Impossible to access feature importance for regressor of type {type(estimator)}. "
f"This method is only valid when the regressor stores internally "
f"the feature importance in the attribute `feature_importances_` "
f"or `coef_`."
)
feature_importance = None
return feature_importance
create_sample_weights(self, series, X_train, y_train_index)
¶
Crate weights for each observation according to the forecaster's attributes
series_weights and weight_func. The resulting weights are product of both
types of weights.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
series |
DataFrame |
Time series used to create |
required |
X_train |
DataFrame |
Dataframe generated with the method |
required |
y_train_index |
Index |
Index of |
required |
Returns:
| Type | Description |
|---|---|
ndarray |
Weights to use in |
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
def create_sample_weights(
self,
series: pd.DataFrame,
X_train: pd.DataFrame,
y_train_index: pd.Index,
)-> np.ndarray:
"""
Crate weights for each observation according to the forecaster's attributes
`series_weights` and `weight_func`. The resulting weights are product of both
types of weights.
Parameters
----------
series : pandas DataFrame
Time series used to create `X_train` with the method `create_train_X_y`.
X_train : pandas DataFrame
Dataframe generated with the method `create_train_X_y`, first return.
y_train_index : pandas Index
Index of `y_train` generated with the method `create_train_X_y`, fourth return.
Returns
-------
weights : numpy ndarray
Weights to use in `fit` method.
"""
weights = None
weights_samples = None
weights_series = None
if self.series_weights is not None:
# Series not present in series_weights have a weight of 1 in all their samples.
# Keys in series_weights not present in series are ignored.
series_not_in_series_weights = set(series.columns) - set(self.series_weights.keys())
if series_not_in_series_weights:
warnings.warn(
f"{series_not_in_series_weights} not present in `series_weights`."
f" A weight of 1 is given to all their samples."
)
self.series_weights_ = dict.fromkeys(series.columns, 1.)
self.series_weights_.update((k, v) for k, v in self.series_weights.items() if k in self.series_weights_)
weights_series = [np.repeat(self.series_weights_[serie], sum(X_train[serie]))
for serie in series.columns]
weights_series = np.concatenate(weights_series)
if self.weight_func is not None:
if isinstance(self.weight_func, Callable):
self.weight_func_ = dict.fromkeys(series.columns, self.weight_func)
else:
# Series not present in weight_func have a weight of 1 in all their samples
series_not_in_weight_func = set(series.columns) - set(self.weight_func.keys())
if series_not_in_weight_func:
warnings.warn(
f"{series_not_in_weight_func} not present in `weight_func`."
f" A weight of 1 is given to all their samples."
)
self.weight_func_ = dict.fromkeys(series.columns, lambda index: np.ones_like(index, dtype=float))
self.weight_func_.update((k, v) for k, v in self.weight_func.items() if k in self.weight_func_)
weights_samples = []
for key in self.weight_func_.keys():
index = y_train_index[X_train[X_train[key] == 1.0].index]
weights_samples.append(self.weight_func_[key](index))
weights_samples = np.concatenate(weights_samples)
if weights_series is not None:
weights = weights_series
if weights_samples is not None:
weights = weights * weights_samples
else:
if weights_samples is not None:
weights = weights_samples
if weights is not None:
if np.isnan(weights).any():
raise ValueError(
"The resulting `weights` cannot have NaN values."
)
if np.any(weights < 0):
raise ValueError(
"The resulting `weights` cannot have negative values."
)
if np.sum(weights) == 0:
raise ValueError(
("The resulting `weights` cannot be normalized because "
"the sum of the weights is zero.")
)
return weights
create_train_X_y(self, series, exog=None)
¶
Create training matrices from multiple time series and exogenous
variables.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
series |
DataFrame |
Training time series. |
required |
exog |
Union[pandas.core.series.Series, pandas.core.frame.DataFrame] |
Exogenous variable/s included as predictor/s. Must have the same
number of observations as |
None |
Returns:
| Type | Description |
|---|---|
Tuple[pandas.core.frame.DataFrame, pandas.core.series.Series, pandas.core.indexes.base.Index, pandas.core.indexes.base.Index] |
Pandas DataFrame with the training values (predictors). |
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
def create_train_X_y(
self,
series: pd.DataFrame,
exog: Optional[Union[pd.Series, pd.DataFrame]]=None
) -> Tuple[pd.DataFrame, pd.Series, pd.Index, pd.Index]:
"""
Create training matrices from multiple time series and exogenous
variables.
Parameters
----------
series : pandas DataFrame
Training time series.
exog : pandas Series, pandas DataFrame, default `None`
Exogenous variable/s included as predictor/s. Must have the same
number of observations as `series` and their indexes must be aligned.
Returns
-------
X_train : pandas DataFrame
Pandas DataFrame with the training values (predictors).
y_train : pandas Series, shape (len(series) - self.max_lag, )
Values (target) of the time series related to each row of `X_train`.
y_index : pandas Index
Index of `series`.
y_train_index: pandas Index
Index of `y_train`.
"""
if not isinstance(series, pd.DataFrame):
raise TypeError(f'`series` must be a pandas DataFrame. Got {type(series)}.')
series_col_names = list(series.columns)
if self.transformer_series is None:
self.transformer_series_ = {serie: None for serie in series_col_names}
elif not isinstance(self.transformer_series, dict):
self.transformer_series_ = {serie: clone(self.transformer_series)
for serie in series_col_names}
else:
self.transformer_series_ = {serie: None for serie in series_col_names}
# Only elements already present in transformer_series_ are updated
self.transformer_series_.update(
(k, v) for k, v in deepcopy(self.transformer_series).items() if k in self.transformer_series_
)
series_not_in_transformer_series = set(series.columns) - set(self.transformer_series.keys())
if series_not_in_transformer_series:
warnings.warn(
f"{series_not_in_transformer_series} not present in `transformer_series`."
f" No transformation is applied to these series."
)
X_levels = []
X_train_col_names = [f"lag_{lag}" for lag in self.lags]
for i, serie in enumerate(series.columns):
y = series[serie]
check_y(y=y)
y = transform_series(
series = y,
transformer = self.transformer_series_[serie],
fit = True,
inverse_transform = False
)
y_values, y_index = preprocess_y(y=y)
X_train_values, y_train_values = self._create_lags(y=y_values)
if i == 0:
X_train = X_train_values
y_train = y_train_values
else:
X_train = np.vstack((X_train, X_train_values))
y_train = np.append(y_train, y_train_values)
X_level = [serie]*len(X_train_values)
X_levels.extend(X_level)
if exog is not None:
if len(exog) != len(series):
raise ValueError(
f'`exog` must have same number of samples as `series`. '
f'length `exog`: ({len(exog)}), length `series`: ({len(series)})'
)
check_exog(exog=exog)
if isinstance(exog, pd.Series):
exog = transform_series(
series = exog,
transformer = self.transformer_exog,
fit = True,
inverse_transform = False
)
else:
exog = transform_dataframe(
df = exog,
transformer = self.transformer_exog,
fit = True,
inverse_transform = False
)
exog_values, exog_index = preprocess_exog(exog=exog)
if not (exog_index[:len(y_index)] == y_index).all():
raise ValueError(
('Different index for `series` and `exog`. They must be equal '
'to ensure the correct alignment of values.')
)
col_names_exog = exog.columns if isinstance(exog, pd.DataFrame) else [exog.name]
X_train_col_names.extend(col_names_exog)
# The first `self.max_lag` positions have to be removed from exog
# since they are not in X_train. Then exog is cloned as many times
# as series.
if exog_values.ndim == 1:
X_train = np.column_stack((
X_train,
np.tile(exog_values[self.max_lag:, ], series.shape[1])
))
else:
X_train = np.column_stack((
X_train,
np.tile(exog_values[self.max_lag:, ], [series.shape[1], 1])
))
X_levels = pd.Series(X_levels)
X_levels = pd.get_dummies(X_levels, dtype=float)
X_train_col_names.extend(X_levels.columns)
X_train = np.column_stack((X_train, X_levels.values))
X_train = pd.DataFrame(
data = X_train,
columns = X_train_col_names
)
y_train = pd.Series(
data = y_train,
name = 'y'
)
y_train_index = pd.Index(
np.tile(
y_index[self.max_lag: ].values,
reps = len(series_col_names)
)
)
self.X_train_col_names = X_train_col_names
return X_train, y_train, y_index, y_train_index
fit(self, series, exog=None, store_in_sample_residuals=True)
¶
Training Forecaster.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
series |
DataFrame |
Training time series. |
required |
exog |
Union[pandas.core.series.Series, pandas.core.frame.DataFrame] |
Exogenous variable/s included as predictor/s. Must have the same
number of observations as |
None |
store_in_sample_residuals |
bool |
if True, in_sample_residuals are stored. |
True |
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
def fit(
self,
series: pd.DataFrame,
exog: Optional[Union[pd.Series, pd.DataFrame]]=None,
store_in_sample_residuals: bool=True
) -> None:
"""
Training Forecaster.
Parameters
----------
series : pandas DataFrame
Training time series.
exog : pandas Series, pandas DataFrame, default `None`
Exogenous variable/s included as predictor/s. Must have the same
number of observations as `series` and their indexes must be aligned so
that series[i] is regressed on exog[i].
store_in_sample_residuals : bool, default `True`
if True, in_sample_residuals are stored.
Returns
-------
None
"""
# Reset values in case the forecaster has already been fitted.
self.index_type = None
self.index_freq = None
self.index_values = None
self.last_window = None
self.included_exog = False
self.exog_type = None
self.exog_col_names = None
self.series_col_names = None
self.X_train_col_names = None
self.in_sample_residuals = None
self.fitted = False
self.training_range = None
self.series_col_names = list(series.columns)
if exog is not None:
self.included_exog = True
self.exog_type = type(exog)
self.exog_col_names = \
exog.columns.to_list() if isinstance(exog, pd.DataFrame) else [exog.name]
if len(set(self.exog_col_names) - set(self.series_col_names)) != len(self.exog_col_names):
raise ValueError(
(f'`exog` cannot contain a column named the same as one of the series'
f' (column names of series).\n'
f' `series` columns : {self.series_col_names}.\n'
f' `exog` columns : {self.exog_col_names}.')
)
X_train, y_train, y_index, y_train_index = self.create_train_X_y(series=series, exog=exog)
sample_weight = self.create_sample_weights(
series = series,
X_train = X_train,
y_train_index = y_train_index,
)
if sample_weight is not None:
if not str(type(self.regressor)) == "<class 'xgboost.sklearn.XGBRegressor'>":
self.regressor.fit(X=X_train, y=y_train, sample_weight=sample_weight)
else:
self.regressor.fit(X=X_train.to_numpy(), y=y_train.to_numpy(), sample_weight=sample_weight)
else:
if not str(type(self.regressor)) == "<class 'xgboost.sklearn.XGBRegressor'>":
self.regressor.fit(X=X_train, y=y_train)
else:
self.regressor.fit(X=X_train.to_numpy(), y=y_train.to_numpy())
self.fitted = True
self.fit_date = pd.Timestamp.today().strftime('%Y-%m-%d %H:%M:%S')
self.training_range = y_index[[0, -1]]
self.index_type = type(y_index)
if isinstance(y_index, pd.DatetimeIndex):
self.index_freq = y_index.freqstr
else:
self.index_freq = y_index.step
self.index_values = y_index
in_sample_residuals = {}
# This is done to save time during fit in functions such as backtesting()
if store_in_sample_residuals:
if not str(type(self.regressor)) == "<class 'xgboost.sklearn.XGBRegressor'>":
residuals = y_train - self.regressor.predict(X_train)
else:
residuals = y_train - self.regressor.predict(X_train.to_numpy())
for serie in series.columns:
in_sample_residuals[serie] = residuals.values[X_train[serie] == 1.]
if len(in_sample_residuals[serie]) > 1000:
# Only up to 1000 residuals are stored
rng = np.random.default_rng(seed=123)
in_sample_residuals[serie] = rng.choice(
a = in_sample_residuals[serie],
size = 1000,
replace = False
)
else:
for serie in series.columns:
in_sample_residuals[serie] = np.array([None])
self.in_sample_residuals = in_sample_residuals
# The last time window of training data is stored so that lags needed as
# predictors in the first iteration of `predict()` can be calculated.
self.last_window = series.iloc[-self.max_lag:, ].copy()
get_feature_importance(self)
¶
Return feature importance of the regressor stored in the
forecaster. Only valid when regressor stores internally the feature
importance in the attribute feature_importances_ or coef_.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
self |
None |
required |
Returns:
| Type | Description |
|---|---|
DataFrame |
Feature importance associated with each predictor. |
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
def get_feature_importance(
self
) -> pd.DataFrame:
"""
Return feature importance of the regressor stored in the
forecaster. Only valid when regressor stores internally the feature
importance in the attribute `feature_importances_` or `coef_`.
Parameters
----------
self
Returns
-------
feature_importance : pandas DataFrame
Feature importance associated with each predictor.
"""
if self.fitted == False:
raise sklearn.exceptions.NotFittedError(
("This forecaster is not fitted yet. Call `fit` with appropriate "
"arguments before using `get_feature_importance()`.")
)
if isinstance(self.regressor, sklearn.pipeline.Pipeline):
estimator = self.regressor[-1]
else:
estimator = self.regressor
try:
feature_importance = pd.DataFrame({
'feature': self.X_train_col_names,
'importance' : estimator.feature_importances_
})
except:
try:
feature_importance = pd.DataFrame({
'feature': self.X_train_col_names,
'importance' : estimator.coef_
})
except:
warnings.warn(
f"Impossible to access feature importance for regressor of type {type(estimator)}. "
f"This method is only valid when the regressor stores internally "
f"the feature importance in the attribute `feature_importances_` "
f"or `coef_`."
)
feature_importance = None
return feature_importance
predict(self, steps, levels=None, last_window=None, exog=None)
¶
Predict n steps ahead. It is an recursive process in which, each prediction,
is used as a predictor for the next step.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
steps |
int |
Number of future steps predicted. |
required |
levels |
Union[str, list] |
Time series to be predicted. If |
None |
last_window |
Optional[pandas.core.frame.DataFrame] |
Values of the series used to create the predictors (lags) need in the first iteration of prediction (t + 1). If |
None |
exog |
Union[pandas.core.series.Series, pandas.core.frame.DataFrame] |
Exogenous variable/s included as predictor/s. |
None |
Returns:
| Type | Description |
|---|---|
DataFrame |
Predicted values, one column for each level. |
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
def predict(
self,
steps: int,
levels: Optional[Union[str, list]]=None,
last_window: Optional[pd.DataFrame]=None,
exog: Optional[Union[pd.Series, pd.DataFrame]]=None
) -> pd.DataFrame:
"""
Predict n steps ahead. It is an recursive process in which, each prediction,
is used as a predictor for the next step.
Parameters
----------
steps : int
Number of future steps predicted.
levels : str, list, default `None`
Time series to be predicted. If `None` all levels will be predicted.
**New in version 0.6.0**
last_window : pandas DataFrame, default `None`
Values of the series used to create the predictors (lags) need in the
first iteration of prediction (t + 1).
If `last_window = None`, the values stored in `self.last_window` are
used to calculate the initial predictors, and the predictions start
right after training data.
exog : pandas Series, pandas DataFrame, default `None`
Exogenous variable/s included as predictor/s.
Returns
-------
predictions : pandas DataFrame
Predicted values, one column for each level.
"""
if levels is None:
levels = self.series_col_names
elif isinstance(levels, str):
levels = [levels]
check_predict_input(
forecaster_type = type(self).__name__,
steps = steps,
fitted = self.fitted,
included_exog = self.included_exog,
index_type = self.index_type,
index_freq = self.index_freq,
window_size = self.window_size,
last_window = last_window,
exog = exog,
exog_type = self.exog_type,
exog_col_names = self.exog_col_names,
interval = None,
max_steps = None,
levels = levels,
series_col_names = self.series_col_names
)
if exog is not None:
if isinstance(exog, pd.DataFrame):
exog = transform_dataframe(
df = exog,
transformer = self.transformer_exog,
fit = False,
inverse_transform = False
)
else:
exog = transform_series(
series = exog,
transformer = self.transformer_exog,
fit = False,
inverse_transform = False
)
exog_values, _ = preprocess_exog(
exog = exog.iloc[:steps, ]
)
else:
exog_values = None
predictions = []
for level in levels:
if last_window is None:
last_window = self.last_window.copy()
last_window_level = transform_series(
series = last_window[level],
transformer = self.transformer_series_[level],
fit = False,
inverse_transform = False
)
last_window_values, last_window_index = preprocess_last_window(
last_window = last_window_level
)
preds_level = self._recursive_predict(
steps = steps,
level = level,
last_window = copy(last_window_values),
exog = copy(exog_values)
)
preds_level = pd.Series(
data = preds_level,
index = expand_index(
index = last_window_index,
steps = steps
),
name = level
)
preds_level = transform_series(
series = preds_level,
transformer = self.transformer_series_[level],
fit = False,
inverse_transform = True
)
predictions.append(preds_level)
predictions = pd.concat(predictions, axis=1)
return predictions
predict_interval(self, steps, levels=None, last_window=None, exog=None, interval=[5, 95], n_boot=500, random_state=123, in_sample_residuals=True)
¶
Iterative process in which, each prediction, is used as a predictor
for the next step and bootstrapping is used to estimate prediction intervals. Both, predictions and intervals, are returned.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
steps |
int |
Number of future steps predicted. |
required |
levels |
Union[str, list] |
Time series to be predicted. If |
None |
last_window |
Optional[pandas.core.frame.DataFrame] |
Values of the series used to create the predictors (lags) needed in the first iteration of prediction (t + 1). If |
None |
exog |
Union[pandas.core.series.Series, pandas.core.frame.DataFrame] |
Exogenous variable/s included as predictor/s. |
None |
interval |
list |
Confidence of the prediction interval estimated. Sequence of
percentiles to compute, which must be between 0 and 100 inclusive.
For example, interval of 95% should be as |
[5, 95] |
n_boot |
int |
Number of bootstrapping iterations used to estimate prediction intervals. |
500 |
random_state |
int |
Sets a seed to the random generator, so that boot intervals are always deterministic. |
123 |
in_sample_residuals |
bool |
If |
True |
Returns:
| Type | Description |
|---|---|
DataFrame |
Values predicted by the forecaster and their estimated interval. level: predictions. level_lower_bound: lower bound of the interval. level_upper_bound: upper bound interval of the interval. |
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
def predict_interval(
self,
steps: int,
levels: Optional[Union[str, list]]=None,
last_window: Optional[pd.DataFrame]=None,
exog: Optional[Union[pd.Series, pd.DataFrame]]=None,
interval: list=[5, 95],
n_boot: int=500,
random_state: int=123,
in_sample_residuals: bool=True
) -> pd.DataFrame:
"""
Iterative process in which, each prediction, is used as a predictor
for the next step and bootstrapping is used to estimate prediction
intervals. Both, predictions and intervals, are returned.
Parameters
----------
steps : int
Number of future steps predicted.
levels : str, list, default `None`
Time series to be predicted. If `None` all levels will be predicted.
**New in version 0.6.0**
last_window : pandas DataFrame, default `None`
Values of the series used to create the predictors (lags) needed in the
first iteration of prediction (t + 1).
If `last_window = None`, the values stored in` self.last_window` are
used to calculate the initial predictors, and the predictions start
right after training data.
exog : pandas Series, pandas DataFrame, default `None`
Exogenous variable/s included as predictor/s.
interval : list, default `[5, 95]`
Confidence of the prediction interval estimated. Sequence of
percentiles to compute, which must be between 0 and 100 inclusive.
For example, interval of 95% should be as `interval = [2.5, 97.5]`.
n_boot : int, default `500`
Number of bootstrapping iterations used to estimate prediction
intervals.
random_state : int, default 123
Sets a seed to the random generator, so that boot intervals are always
deterministic.
in_sample_residuals : bool, default `True`
If `True`, residuals from the training data are used as proxy of
prediction error to create prediction intervals. If `False`, out of
sample residuals are used. In the latter case, the user should have
calculated and stored the residuals within the forecaster (see
`set_out_sample_residuals()`).
Returns
-------
predictions : pandas DataFrame
Values predicted by the forecaster and their estimated interval.
level: predictions.
level_lower_bound: lower bound of the interval.
level_upper_bound: upper bound interval of the interval.
Notes
-----
More information about prediction intervals in forecasting:
https://otexts.com/fpp2/prediction-intervals.html
Forecasting: Principles and Practice (2nd ed) Rob J Hyndman and
George Athanasopoulos.
"""
if levels is None:
levels = self.series_col_names
elif isinstance(levels, str):
levels = [levels]
for level in levels:
if in_sample_residuals and (self.in_sample_residuals[level] == None).any():
raise ValueError(
(f"`forecaster.in_sample_residuals['{level}']` contains `None` values. "
"Try using `fit` method with `in_sample_residuals=True` or set in "
"`predict_interval` method `in_sample_residuals=False` and use "
"`out_sample_residuals` (see `set_out_sample_residuals()`).")
)
if not in_sample_residuals and self.out_sample_residuals is None:
raise ValueError(
('`forecaster.out_sample_residuals` is `None`. Use '
'`in_sample_residuals=True` or method `set_out_sample_residuals()` '
'before `predict_interval()`.')
)
if not in_sample_residuals and len(set(levels) - set(self.out_sample_residuals.keys())) != 0:
raise ValueError(
(f'Not `forecaster.out_sample_residuals` for levels: {set(levels) - set(self.out_sample_residuals.keys())}. '
f'Use method `set_out_sample_residuals()`.')
)
check_predict_input(
forecaster_type = type(self).__name__,
steps = steps,
fitted = self.fitted,
included_exog = self.included_exog,
index_type = self.index_type,
index_freq = self.index_freq,
window_size = self.window_size,
last_window = last_window,
exog = exog,
exog_type = self.exog_type,
exog_col_names = self.exog_col_names,
interval = interval,
max_steps = None,
levels = levels,
series_col_names = self.series_col_names
)
if exog is not None:
if isinstance(exog, pd.DataFrame):
exog = transform_dataframe(
df = exog,
transformer = self.transformer_exog,
fit = False,
inverse_transform = False
)
else:
exog = transform_series(
series = exog,
transformer = self.transformer_exog,
fit = False,
inverse_transform = False
)
exog_values, _ = preprocess_exog(
exog = exog.iloc[:steps, ]
)
else:
exog_values = None
predictions = []
for level in levels:
if last_window is None:
last_window = self.last_window.copy()
last_window_level = transform_series(
series = last_window[level],
transformer = self.transformer_series_[level],
fit = False,
inverse_transform = False
)
last_window_values, last_window_index = preprocess_last_window(
last_window = last_window_level
)
# Since during predict() `last_window_values` and `exog_values` are modified,
# the originals are stored to be used later.
last_window_values_original = last_window_values.copy()
if exog is not None:
exog_values_original = exog_values.copy()
else:
exog_values_original = None
preds_level = self._recursive_predict(
steps = steps,
level = level,
last_window = last_window_values,
exog = exog_values
)
preds_level_interval = self._estimate_boot_interval(
steps = steps,
level = level,
last_window = copy(last_window_values_original),
exog = copy(exog_values_original),
interval = interval,
n_boot = n_boot,
random_state = random_state,
in_sample_residuals = in_sample_residuals
)
preds_level = np.column_stack((preds_level, preds_level_interval))
preds_level = pd.DataFrame(
data = preds_level,
index = expand_index(
index = last_window_index,
steps = steps
),
columns = [level, f'{level}_lower_bound', f'{level}_upper_bound']
)
if self.transformer_series_[level]:
for col in preds_level.columns:
preds_level[col] = self.transformer_series_[level].inverse_transform(preds_level[[col]])
predictions.append(preds_level)
predictions = pd.concat(predictions, axis=1)
return predictions
set_lags(self, lags)
¶
Set new value to the attribute lags.
Attributes max_lag and window_size are also updated.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
lags |
Union[int, list, numpy.ndarray, range] |
Lags used as predictors. Index starts at 1, so lag 1 is equal to t-1.
|
required |
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
def set_lags(
self,
lags: Union[int, list, np.ndarray, range]
) -> None:
"""
Set new value to the attribute `lags`.
Attributes `max_lag` and `window_size` are also updated.
Parameters
----------
lags : int, list, 1D np.array, range
Lags used as predictors. Index starts at 1, so lag 1 is equal to t-1.
`int`: include lags from 1 to `lags`.
`list` or `np.array`: include only lags present in `lags`.
Returns
-------
None
"""
self.lags = initialize_lags(type(self).__name__, lags)
self.max_lag = max(self.lags)
self.window_size = max(self.lags)
set_out_sample_residuals(self, residuals, append=True, transform=True, random_state=123)
¶
Set new values to the attribute out_sample_residuals. Out of sample
residuals are meant to be calculated using observations that did not participate in the training process.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
residuals |
DataFrame |
Values of residuals. If len(residuals) > 1000, only a random sample
of 1000 values are stored. Columns must be the same as |
required |
append |
bool |
If |
True |
transform |
bool |
If |
True |
random_state |
int |
Sets a seed to the random sampling for reproducible output. |
123 |
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
def set_out_sample_residuals(
self,
residuals: pd.DataFrame,
append: bool=True,
transform: bool=True,
random_state: int=123
)-> None:
"""
Set new values to the attribute `out_sample_residuals`. Out of sample
residuals are meant to be calculated using observations that did not
participate in the training process.
Parameters
----------
residuals : pandas DataFrame
Values of residuals. If len(residuals) > 1000, only a random sample
of 1000 values are stored. Columns must be the same as `levels`.
append : bool, default `True`
If `True`, new residuals are added to the once already stored in the
attribute `out_sample_residuals`. Once the limit of 1000 values is
reached, no more values are appended. If False, `out_sample_residuals`
is overwritten with the new residuals.
transform : bool, default `True`
If `True`, new residuals are transformed using self.transformer_series.
random_state : int, default `123`
Sets a seed to the random sampling for reproducible output.
Returns
-------
self
"""
if not isinstance(residuals, pd.DataFrame):
raise TypeError(
f"`residuals` argument must be a pandas DataFrame. Got {type(residuals)}."
)
if not self.fitted:
raise sklearn.exceptions.NotFittedError(
("This forecaster is not fitted yet. Call `fit` with appropriate "
"arguments before using `set_out_sample_residuals()`.")
)
out_sample_residuals = {}
for level in residuals.columns:
if not level in self.series_col_names:
continue
else:
residuals_level = residuals[level]
if not transform and self.transformer_series_[level] is not None:
warnings.warn(
('Argument `transform` is set to `False` but forecaster was trained '
f'using a transformer {self.transformer_series_[level]} for level {level}. '
'Ensure that the new residuals are already transformed or set `transform=True`.')
)
if transform and self.transformer_series_ and self.transformer_series_[level]:
warnings.warn(
('Residuals will be transformed using the same transformer used '
f'when training the forecaster for level {level} : ({self.transformer_series_[level]}). '
'Ensure that the new residuals are on the same scale as the '
'original time series. ')
)
residuals_level = transform_series(
series = residuals_level,
transformer = self.transformer_series_[level],
fit = False,
inverse_transform = False
)
if len(residuals_level) > 1000:
rng = np.random.default_rng(seed=random_state)
residuals_level = rng.choice(a=residuals_level, size=1000, replace=False)
if append and self.out_sample_residuals is not None:
if not level in self.out_sample_residuals.keys():
raise ValueError(
f'{level} does not exists in `forecaster.out_sample_residuals` keys: {list(self.out_sample_residuals.keys())}'
)
free_space = max(0, 1000 - len(self.out_sample_residuals[level]))
if len(residuals_level) < free_space:
residuals_level = np.hstack((
self.out_sample_residuals[level],
residuals_level
))
else:
residuals_level = np.hstack((
self.out_sample_residuals[level],
residuals_level[:free_space]
))
out_sample_residuals[level] = np.array(residuals_level)
self.out_sample_residuals = out_sample_residuals
set_params(self, **params)
¶
Set new values to the parameters of the scikit learn model stored in the
forecaster.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
params |
dict |
Parameters values. |
{} |
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
def set_params(
self,
**params: dict
) -> None:
"""
Set new values to the parameters of the scikit learn model stored in the
forecaster.
Parameters
----------
params : dict
Parameters values.
Returns
-------
self
"""
self.regressor = clone(self.regressor)
self.regressor.set_params(**params)
_create_lags(self, y)
private
¶
Transforms a 1d array into a 2d array (X) and a 1d array (y). Each row
in X is associated with a value of y and it represents the lags that precede it.
Notice that, the returned matrix X_data, contains the lag 1 in the first column, the lag 2 in the second column and so on.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
y |
ndarray |
Training time series. |
required |
Returns:
| Type | Description |
|---|---|
Tuple[numpy.ndarray, numpy.ndarray] |
2d numpy array with the lagged values (predictors). |
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
def _create_lags(
self,
y: np.ndarray
) -> Tuple[np.ndarray, np.ndarray]:
"""
Transforms a 1d array into a 2d array (X) and a 1d array (y). Each row
in X is associated with a value of y and it represents the lags that
precede it.
Notice that, the returned matrix X_data, contains the lag 1 in the first
column, the lag 2 in the second column and so on.
Parameters
----------
y : 1d numpy ndarray
Training time series.
Returns
-------
X_data : 2d numpy ndarray, shape (samples - max(self.lags), len(self.lags))
2d numpy array with the lagged values (predictors).
y_data : 1d numpy ndarray, shape (samples - max(self.lags),)
Values of the time series related to each row of `X_data`.
"""
n_splits = len(y) - self.max_lag
if n_splits <= 0:
raise ValueError(
f'The maximum lag ({self.max_lag}) must be less than the length '
f'of the series ({len(y)}).'
)
X_data = np.full(shape=(n_splits, len(self.lags)), fill_value=np.nan, dtype=float)
for i, lag in enumerate(self.lags):
X_data[:, i] = y[self.max_lag - lag: -lag]
y_data = y[self.max_lag:]
return X_data, y_data
_estimate_boot_interval(self, steps, level, last_window=None, exog=None, interval=[5, 95], n_boot=500, random_state=123, in_sample_residuals=True)
private
¶
Iterative process in which, each prediction, is used as a predictor
for the next step and bootstrapping is used to estimate prediction intervals. This method only returns prediction intervals. See predict_intervals() to calculate both, predictions and intervals.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
steps |
int |
Number of future steps predicted. |
required |
level |
str |
Time series to be predicted. |
required |
last_window |
Optional[numpy.ndarray] |
Values of the series used to create the predictors (lags) needed in the first iteration of prediction (t + 1). If |
None |
exog |
Optional[numpy.ndarray] |
Exogenous variable/s included as predictor/s. |
None |
n_boot |
int |
Number of bootstrapping iterations used to estimate prediction intervals. |
500 |
random_state |
int |
Sets a seed to the random generator, so that boot intervals are always deterministic. |
123 |
interval |
list |
Confidence of the prediction interval estimated. Sequence of
percentiles to compute, which must be between 0 and 100 inclusive.
For example, interval of 95% should be as |
[5, 95] |
in_sample_residuals |
bool |
If |
True |
Returns:
| Type | Description |
|---|---|
ndarray |
Interval estimated for each prediction by bootstrapping: lower_bound: lower bound of the interval. upper_bound: upper bound interval of the interval. |
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
def _estimate_boot_interval(
self,
steps: int,
level: str,
last_window: Optional[np.ndarray]=None,
exog: Optional[np.ndarray]=None,
interval: list=[5, 95],
n_boot: int=500,
random_state: int=123,
in_sample_residuals: bool=True
) -> np.ndarray:
"""
Iterative process in which, each prediction, is used as a predictor
for the next step and bootstrapping is used to estimate prediction
intervals. This method only returns prediction intervals.
See predict_intervals() to calculate both, predictions and intervals.
Parameters
----------
steps : int
Number of future steps predicted.
level : str
Time series to be predicted.
last_window : 1d numpy ndarray shape (, max_lag), default `None`
Values of the series used to create the predictors (lags) needed in the
first iteration of prediction (t + 1).
If `last_window = `None`, the values stored in` self.last_window` are
used to calculate the initial predictors, and the predictions start
right after training data.
exog : numpy ndarray, default `None`
Exogenous variable/s included as predictor/s.
n_boot : int, default `500`
Number of bootstrapping iterations used to estimate prediction
intervals.
random_state : int
Sets a seed to the random generator, so that boot intervals are always
deterministic.
interval : list, default `[5, 95]`
Confidence of the prediction interval estimated. Sequence of
percentiles to compute, which must be between 0 and 100 inclusive.
For example, interval of 95% should be as `interval = [2.5, 97.5]`.
in_sample_residuals : bool, default `True`
If `True`, residuals from the training data are used as proxy of
prediction error to create prediction intervals. If `False`, out of
sample residuals are used. In the latter case, the user should have
calculated and stored the residuals within the forecaster (see
`set_out_sample_residuals()`).
Returns
-------
prediction_interval : numpy ndarray, shape (steps, 2)
Interval estimated for each prediction by bootstrapping:
lower_bound: lower bound of the interval.
upper_bound: upper bound interval of the interval.
Notes
-----
More information about prediction intervals in forecasting:
https://otexts.com/fpp2/prediction-intervals.html
Forecasting: Principles and Practice (2nd ed) Rob J Hyndman and
George Athanasopoulos.
"""
if last_window is None:
last_window = self.last_window[level]
last_window = last_window.values
boot_predictions = np.full(
shape = (steps, n_boot),
fill_value = np.nan,
dtype = float
)
rng = np.random.default_rng(seed=random_state)
seeds = rng.integers(low=0, high=10000, size=n_boot)
for i in range(n_boot):
# In each bootstraping iteration the initial last_window and exog
# need to be restored.
last_window_boot = last_window.copy()
if exog is not None:
exog_boot = exog.copy()
else:
exog_boot = None
if in_sample_residuals:
residuals = self.in_sample_residuals[level]
else:
residuals = self.out_sample_residuals[level]
rng = np.random.default_rng(seed=seeds[i])
sample_residuals = rng.choice(
a = residuals,
size = steps,
replace = True
)
for step in range(steps):
prediction = self._recursive_predict(
steps = 1,
level = level,
last_window = last_window_boot,
exog = exog_boot
)
prediction_with_residual = prediction + sample_residuals[step]
boot_predictions[step, i] = prediction_with_residual
last_window_boot = np.append(
last_window_boot[1:],
prediction_with_residual
)
if exog is not None:
exog_boot = exog_boot[1:]
prediction_interval = np.percentile(boot_predictions, q=interval, axis=1)
prediction_interval = prediction_interval.transpose()
return prediction_interval
_recursive_predict(self, steps, level, last_window, exog=None)
private
¶
Predict n steps ahead. It is an iterative process in which, each prediction,
is used as a predictor for the next step.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
steps |
int |
Number of future steps predicted. |
required |
level |
str |
Time series to be predicted. |
required |
last_window |
ndarray |
Values of the series used to create the predictors (lags) need in the first iteration of prediction (t + 1). |
required |
exog |
Optional[numpy.ndarray] |
Exogenous variable/s included as predictor/s. |
None |
Returns:
| Type | Description |
|---|---|
ndarray |
Predicted values. |
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
def _recursive_predict(
self,
steps: int,
level: str,
last_window: np.ndarray,
exog: Optional[np.ndarray]=None
) -> np.ndarray:
"""
Predict n steps ahead. It is an iterative process in which, each prediction,
is used as a predictor for the next step.
Parameters
----------
steps : int
Number of future steps predicted.
level : str
Time series to be predicted.
last_window : numpy ndarray
Values of the series used to create the predictors (lags) need in the
first iteration of prediction (t + 1).
exog : numpy ndarray, default `None`
Exogenous variable/s included as predictor/s.
Returns
-------
predictions : numpy ndarray
Predicted values.
"""
predictions = np.full(shape=steps, fill_value=np.nan)
for i in range(steps):
X = last_window[-self.lags].reshape(1, -1)
if exog is not None:
X = np.column_stack((X, exog[i, ].reshape(1, -1)))
levels_dummies = np.zeros(shape=(1, len(self.series_col_names)), dtype=float)
levels_dummies[0][self.series_col_names.index(level)] = 1.
X = np.column_stack((X, levels_dummies.reshape(1, -1)))
with warnings.catch_warnings():
# Suppress scikit-learn warning: "X does not have valid feature names,
# but NoOpTransformer was fitted with feature names".
warnings.simplefilter("ignore")
prediction = self.regressor.predict(X)
predictions[i] = prediction.ravel()[0]
# Update `last_window` values. The first position is discarded and
# the new prediction is added at the end.
last_window = np.append(last_window[1:], prediction)
return predictions
__repr__(self)
special
¶
Information displayed when a ForecasterAutoregMultiSeries object is printed.
Source code in skforecast/ForecasterAutoregMultiSeries/ForecasterAutoregMultiSeries.py
def __repr__(
self
) -> str:
"""
Information displayed when a ForecasterAutoregMultiSeries object is printed.
"""
if isinstance(self.regressor, sklearn.pipeline.Pipeline):
name_pipe_steps = tuple(name + "__" for name in self.regressor.named_steps.keys())
params = {key : value for key, value in self.regressor.get_params().items() \
if key.startswith(name_pipe_steps)}
else:
params = self.regressor.get_params()
info = (
f"{'=' * len(str(type(self)).split('.')[1])} \n"
f"{str(type(self)).split('.')[1]} \n"
f"{'=' * len(str(type(self)).split('.')[1])} \n"
f"Regressor: {self.regressor} \n"
f"Lags: {self.lags} \n"
f"Transformer for series: {self.transformer_series} \n"
f"Transformer for exog: {self.transformer_exog} \n"
f"Window size: {self.window_size} \n"
f"Series levels (names): {self.series_col_names} \n"
f"Series weights: {self.series_weights} \n"
f"Weight function included: {True if self.weight_func is not None else False} \n"
f"Exogenous included: {self.included_exog} \n"
f"Type of exogenous variable: {self.exog_type} \n"
f"Exogenous variables names: {self.exog_col_names} \n"
f"Training range: {self.training_range.to_list() if self.fitted else None} \n"
f"Training index type: {str(self.index_type).split('.')[-1][:-2] if self.fitted else None} \n"
f"Training index frequency: {self.index_freq if self.fitted else None} \n"
f"Regressor parameters: {params} \n"
f"Creation date: {self.creation_date} \n"
f"Last fit date: {self.fit_date} \n"
f"Skforecast version: {self.skforcast_version} \n"
f"Python version: {self.python_version} \n"
)
return info