Back to models
Transformer

TimeBinAggregate

API referenceSource
MultivariateMissing valuesUnequal lengthUnequal length removes

Bins time series and aggregates by bin.

In transform, applies groupby with aggfunc on the temporal coordinate.

More precisely: bins encodes bins \(B_1, \dots, B_k\) where \(B_i\) are intervals, in the reals or in a temporal (time stamp) range.

In transform, the estimator TimeBinAggregate collects values at time stamps of X falling into \(B_i\) as a sample \(S_i\), and then applies aggfunc to \(S_i\) to obtain an aggregate value \(v_i\). The transformed series are values \(v_i\) at time stamps \(t_i\), determined from \(B_i\) per the rule in return_index.

Quickstart

python
from sktime.transformations.binning import TimeBinAggregate

estimator = TimeBinAggregate(bins, aggfunc=None, return_index='bin_start')

Parameters(3)

bins1D array-like or pd.IntervalIndex
if 1D array-like, is interpreted as breaks of bins breaks of bins defining intervals considered by aggfunc
aggfunccallable * 1D array-like -> float), optional, default=np.mean
Function used to aggregate the values in intervals. Should have signature 1D -> float and defaults to mean if None
return_indexstr, one of the below; optional, default=”range”

  • "range" = RangeIndex with bins indexed in same order as in bins

  • "bin_start" = transformed pd.DataFrame will be indexed by bin starts

  • "bin_end" = transformed pd.DataFrame will be indexed by bin ends

  • "bin_mid" = transformed pd.DataFrame will be indexed by bin midpoints

  • "bin" = transformed pd.DataFrame will have bins as IntervalIndex

Examples

Basic usage:
>>> import pandas as pd
>>> from sktime.transformations.binning import TimeBinAggregate
>>> X = pd. DataFrame (
... { "a": [1, 2, 3, 4, 5, 6, 7, 8, 9, 10 ]}, index = list (range (10))
... )
>>> t = TimeBinAggregate (bins = [- 1, 2, 5, 10 ])
>>> t. fit_transform (X) 0 -1 2.0 2 5.0 5 8.5 Using return_index="range" to get a plain sequential index instead of bin starts:
>>> t2 = TimeBinAggregate (bins = [- 1, 2, 5, 10 ], return_index = "range")
>>> t2. fit_transform (X) 0 0 2.0 1 5.0 2 8.5 Using return_index="bin_end" to index by the end of each bin instead:
>>> t3 = TimeBinAggregate (bins = [- 1, 2, 5, 10 ], return_index = "bin_end")
>>> t3. fit_transform (X) 0 2 2.0 5 5.0 10 8.5