Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Remove step_length hyper-parameter from reduction classes #900

Merged
merged 5 commits into from Jun 3, 2021
Merged

Remove step_length hyper-parameter from reduction classes #900

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
2449cd9
Remove step length argument
mloning May 21, 2021
5d383ff
Merge branch 'main' into fix-reduce-step-length
mloning Jun 2, 2021
4f6dd8a
Remove step length argument
mloning May 21, 2021
7477649
Fix docstrings
mloning Jun 2, 2021
0c81ccc
Merge branch 'fix-reduce-step-length' of https://github.com/alan-turi…
mloning Jun 2, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
75 changes: 31 additions & 44 deletions sktime/forecasting/compose/_reduce.py
View file
Open in desktop
Expand Up @@ -2,6 +2,7 @@
# -*- coding: utf-8 -*-
# copyright: sktime developers, BSD-3-Clause License (see LICENSE file)

"""Composition functionality for reduction approaches to forecasting."""

__author__ = [
"Ayushmaan Seth",
Expand Down Expand Up @@ -37,12 +38,10 @@
from sktime.regression.base import BaseRegressor
from sktime.utils._maint import deprecated
from sktime.utils.validation import check_window_length
from sktime.utils.validation.forecasting import check_step_length


def _concat_y_X(y, X):
"""Helper function to concatenate y and X in preparation for sliding-window
transform"""
"""Concatenate y and X prior to sliding-window transform."""
z = y.to_numpy()
if z.ndim == 1:
z = z.reshape(-1, 1)
Expand All @@ -52,7 +51,7 @@ def _concat_y_X(y, X):


def _check_fh(fh):
"""Helper function to check fh for sliding-window transform"""
"""Check fh prior to sliding-window transform."""
assert fh.is_relative
assert fh.is_all_out_of_sample()
return fh.to_indexer().to_numpy()
Expand All @@ -61,7 +60,7 @@ def _check_fh(fh):
def _sliding_window_transform(
y, window_length, fh, X=None, scitype="tabular-regressor"
):
"""Transform time series data `y` and `X` using sliding window.
"""Transform time series data using sliding window.

See `test_sliding_window_transform_explicit` in test_reduce.py for explicit
example.
Expand Down Expand Up @@ -143,15 +142,13 @@ def _sliding_window_transform(


class _Reducer(_BaseWindowForecaster):
"""Base class for reducing forecasting to time series regression"""
"""Base class for reducing forecasting to regression."""

_required_parameters = ["estimator"]

def __init__(self, estimator, window_length=10, step_length=1):
def __init__(self, estimator, window_length=10):
super(_Reducer, self).__init__(window_length=window_length)
self.estimator = estimator
self.step_length = step_length
self.step_length_ = None
self._cv = None

def fit(self, y, X=None, fh=None):
Expand All @@ -168,13 +165,13 @@ def fit(self, y, X=None, fh=None):

Returns
-------
self : returns an instance of self.
self : Estimator
An fitted instance of self.
"""
n_timepoints = len(y)
self._set_y_X(y, X)
self._set_fh(fh)

self.step_length_ = check_step_length(self.step_length)
self.window_length_ = check_window_length(self.window_length, n_timepoints)

self._fit(y, X)
Expand All @@ -185,8 +182,7 @@ def _fit(self, y, X):
raise NotImplementedError("abstract method")

def _is_predictable(self, last_window):
"""Helper function to check if we can make predictions from last
window"""
"""Check if we can make predictions from last window."""
return (
len(last_window) == self.window_length_
and np.sum(np.isnan(last_window)) == 0
Expand Down Expand Up @@ -223,13 +219,13 @@ def _fit(self, y, X=None):
----------
y : pd.Series
Target time series to which to fit the forecaster.
fh : int, list or np.array, optional (default=None)
The forecasters horizon with the steps ahead to to predict.
X : pd.DataFrame, optional (default=None)
Exogenous variables are ignored

Returns
-------
self : returns an instance of self.
self : Estimator
An fitted instance of self.
"""
# We currently only support out-of-sample predictions. For the direct
# strategy, we need to check this at the beginning of fit, as the fh is
Expand Down Expand Up @@ -465,9 +461,11 @@ def _fit(self, y, X=None, fh=None):
The forecasters horizon with the steps ahead to to predict.
X : pd.DataFrame, optional (default=None)
For this estimator, exogenous variables are ignored

Returns
-------
self : returns an instance of self.
self : Estimator
An fitted instance of self.
"""
# Exogenous variables are not yet support for the dirrec strategy.
if X is not None:
Expand Down Expand Up @@ -555,8 +553,7 @@ def _predict_last_window(

class DirectTabularRegressionForecaster(_DirectReducer):
"""
Forecasting based on reduction to tabular regression using the direct
strategy.
Direct reduction from forecasting to tabular regression.

For the direct reduction strategy, a separate forecaster is fitted
for each step ahead of the forecasting horizon.
Expand All @@ -575,8 +572,7 @@ class DirectTabularRegressionForecaster(_DirectReducer):

class MultioutputTabularRegressionForecaster(_MultioutputReducer):
"""
Forecasting based on reduction to tabular regression using the multioutput
strategy.
Multioutput reduction from forecasting to tabular regression.

For the multioutput strategy, a single estimator capable of handling multioutput
targets is fitted to all the future steps in the forecasting horizon.
Expand All @@ -595,8 +591,7 @@ class MultioutputTabularRegressionForecaster(_MultioutputReducer):

class RecursiveTabularRegressionForecaster(_RecursiveReducer):
"""
Forecasting based on reduction to tabular regression using the recursive
strategy.
Recursive reduction from forecasting to tabular regression.

For the recursive strategy, a single estimator is fit for a one-step-ahead
forecasting horizon and then called iteratively to predict multiple steps ahead.
Expand All @@ -615,10 +610,9 @@ class RecursiveTabularRegressionForecaster(_RecursiveReducer):

class DirRecTabularRegressionForecaster(_DirRecReducer):
"""
Forecasting based on reduction to tabular regression with the
dirrec (hybrid) strategy.
Dir-rec reduction from forecasting to tabular regression.

For the dirrec strategy, a separate forecaster is fitted
For the hybrid dir-rec strategy, a separate forecaster is fitted
for each step ahead of the forecasting horizon and then
the previous forecasting horizon is added as an input
for training the next forecaster, following the recusrive
Expand All @@ -638,8 +632,7 @@ class DirRecTabularRegressionForecaster(_DirRecReducer):

class DirectTimeSeriesRegressionForecaster(_DirectReducer):
"""
Forecasting based on reduction to time-series regression using the direct
strategy.
Direct reduction from forecasting to time-series regression.

For the direct reduction strategy, a separate forecaster is fitted
for each step ahead of the forecasting horizon.
Expand All @@ -658,8 +651,7 @@ class DirectTimeSeriesRegressionForecaster(_DirectReducer):

class MultioutputTimeSeriesRegressionForecaster(_MultioutputReducer):
"""
Forecasting based on reduction to time series regression with a multioutput
reduction strategy.
Multioutput reduction from forecasting to time series regression.

For the multioutput strategy, a single estimator capable of handling multioutput
targets is fitted to all the future steps in the forecasting horizon.
Expand All @@ -678,8 +670,7 @@ class MultioutputTimeSeriesRegressionForecaster(_MultioutputReducer):

class RecursiveTimeSeriesRegressionForecaster(_RecursiveReducer):
"""
Forecasting based on reduction to time series regression with a recursive
reduction strategy.
Recursive reduction from forecasting to time series regression.

For the recursive strategy, a single estimator is fit for a one-step-ahead
forecasting horizon and then called iteratively to predict multiple steps ahead.
Expand All @@ -698,10 +689,9 @@ class RecursiveTimeSeriesRegressionForecaster(_RecursiveReducer):

class DirRecTimeSeriesRegressionForecaster(_DirRecReducer):
"""
Forecasting based on reduction to time-series regression with the
dirrec (hybrid) strategy.
Dir-rec reduction from forecasting to time-series regression.

For the dirrec strategy, a separate forecaster is fitted
For the hybrid dir-rec strategy, a separate forecaster is fitted
for each step ahead of the forecasting horizon and then
the previous forecasting horizon is added as an input
for training the next forecaster, following the recusrive
Expand All @@ -724,7 +714,7 @@ def ReducedForecaster(
estimator, scitype="infer", strategy="recursive", window_length=10, step_length=1
):
"""
Forecasting based on reduction.
Reduction from forecasting to tabular or time series regression.

During fitting, a sliding-window approach is used to first transform the
time series into tabular or panel data, which is then used to fit a tabular or
Expand All @@ -733,7 +723,8 @@ def ReducedForecaster(

Parameters
----------
estimator : a estimator of type given by parameter scitype
estimator : Estimator
A estimator of type given by parameter scitype
scitype : str {"infer", "tabular-regressor", "time-series-regressor"}
Scitype of estimator.
strategy : str {"recursive", "direct", "multioutput"}, optional
Expand Down Expand Up @@ -761,7 +752,7 @@ def ReducedRegressionForecaster(
estimator, scitype, strategy="recursive", window_length=10, step_length=1
):
"""
Forecasting based on reduction.
Reduction from forecasting to tabular or time series regression.

During fitting, a sliding-window approach is used to first transform the
time series into tabular or panel data, which is then used to fit a tabular or
Expand Down Expand Up @@ -800,8 +791,7 @@ def make_reduction(
scitype="infer",
):
"""
Create a reduction forecaster based on a tabular or time-series regression
estimator.
Make forecaster based on reduction to tabular or time-series regression.

During fitting, a sliding-window approach is used to first transform the
time series into tabular or panel data, which is then used to fit a tabular or
Expand Down Expand Up @@ -880,10 +870,7 @@ def _check_strategy(strategy):


def _get_forecaster(scitype, strategy):
"""Helper function to select forecaster for a given scientific type (
scitype)
and reduction strategy"""

"""Select forecaster for a given scientific type and reduction strategy."""
registry = {
"tabular-regressor": {
"direct": DirectTabularRegressionForecaster,
Expand Down