preprocessing.imputation.WeightFunction

preprocessing.imputation.WeightFunction(weights_series)

Callable class for sample weights that can be pickled.

This class wraps the weights_series and provides a callable interface compatible with ForecasterRecursive’s weight_func parameter. Unlike local functions with closures, instances of this class can be pickled using standard pickle/joblib.

When all weights for the requested index sum to zero (e.g. the entire training window falls inside gap-penalty zones after outlier detection), __call__ returns None instead of an all-zero array. ForecasterRecursive.create_sample_weights treats a None return as “no weighting”, avoiding a ValueError while still applying the weighted imputation for windows that do contain positive weights.

Parameters

Name	Type	Description	Default
weights_series	pd.Series	Series containing weight values for each index.	required

Examples

import pickle
import numpy as np
import pandas as pd
from spotforecast2_safe.preprocessing.imputation import WeightFunction

weights = pd.Series([1.0, 0.9, 0.8], index=[0, 1, 2])
weight_func = WeightFunction(weights)

# Callable interface: returns weights for a pd.Index
result = weight_func(pd.Index([0, 1]))
print(result)
assert np.allclose(result, [1.0, 0.9])

# Returns None when all weights in the window are zero
zero_weights = pd.Series([0.0, 0.0, 0.0], index=[0, 1, 2])
wf_zero = WeightFunction(zero_weights)
assert wf_zero(pd.Index([0, 1])) is None

# Can be pickled and unpickled (no closure, fully serialisable)
unpickled = pickle.loads(pickle.dumps(weight_func))
result2 = unpickled(pd.Index([0, 1]))
print(result2)
assert np.allclose(result2, [1.0, 0.9])

WeightFunction: all sample weights for the requested index are zero (the window falls entirely within gap-penalty zones). Returning None so ForecasterRecursive uses uniform weighting.

[1.  0.9]
[1.  0.9]