Unexpected KeyError when using fh is longer than data size in make_reduction

[yarnabrina]

import warnings

import numpy
import pandas
from sklearn.linear_model import LinearRegression
from sklearn.preprocessing import StandardScaler

from sktime.forecasting.base import ForecastingHorizon
from sktime.forecasting.compose import (
    ForecastingPipeline,
    ForecastX,
    TransformedTargetForecaster,
    make_reduction,
)
from sktime.forecasting.model_selection import temporal_train_test_split
from sktime.transformations.compose import TransformerPipeline
from sktime.transformations.series.adapt import TabularToSeriesAdaptor
from sktime.transformations.series.detrend import Deseasonalizer, Detrender
from sktime.transformations.series.impute import Imputer

warnings.filterwarnings("ignore")

df = pandas.DataFrame(
    {
        "abcd": numpy.random.random(size=40),
        "efgh": numpy.random.random(size=40),
        "ijkl": numpy.random.random(size=40),
        "mnop": numpy.random.random(size=40),
        "qrst": 2 * pandas.date_range(start="2024-01-01", end="2024-01-20", freq="1D").tolist(),
        "uv": ["wx"] * 20 + ["yz"] * 20,
    }
)
df = df.set_index(["uv", "qrst"])

y = df[["efgh"]]
X = df[["mnop", "abcd", "ijkl"]]

fh = ForecastingHorizon(range(1, 20 + 1))

sklearn_scaler = StandardScaler()
sktime_scaler = TabularToSeriesAdaptor(sklearn_scaler)

sktime_imputer = Imputer(method="median")

sktime_detrender = Detrender()
sktime_deseasonaliser = Deseasonalizer()

X_transformer = TransformerPipeline([("scaler", sktime_scaler), ("imputer", sktime_imputer)])
y_transformer = TransformerPipeline(
    [("detrender", sktime_detrender), ("deseasonaliser", sktime_deseasonaliser)]
)

sklearn_lm = LinearRegression()
sktime_lm = make_reduction(sklearn_lm, pooling="global", windows_identical=False)
model_lm = TransformedTargetForecaster(
    [("y_transformer", y_transformer), ("reduction_lm", sktime_lm)]
)

forecaster_lm = ForecastX(model_lm.clone(), model_lm.clone())
pipeline_lm = ForecastingPipeline([("X_transformer", X_transformer), ("model_lm", forecaster_lm)])

y_past, _, X_past, _ = temporal_train_test_split(y, X=X, test_size=max(fh))
pipeline_lm.fit(y_past, X=X_past, fh=fh)

This fails with following error:

Traceback (most recent call last):
  File "/home/anirban/conda-environments/sktime/lib/python3.10/site-packages/pandas/core/indexes/range.py", line 345, in get_loc
    return self._range.index(new_key)
ValueError: 1 is not in range

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/anirban/sktime-fork/sktime/forecasting/base/_base.py", line 358, in fit
    self._fit(y=y_inner, X=X_inner, fh=fh)
  File "/home/anirban/sktime-fork/sktime/forecasting/compose/_pipeline.py", line 485, in _fit
    X = t.fit_transform(X=X, y=y)
  File "/home/anirban/sktime-fork/sktime/transformations/base.py", line 629, in fit_transform
    return self.fit(X, y).transform(X, y)
  File "/home/anirban/sktime-fork/sktime/transformations/base.py", line 446, in fit
    self._fit(X=X_inner, y=y_inner)
  File "/home/anirban/sktime-fork/sktime/transformations/compose/_pipeline.py", line 285, in _fit
    Xt = transformer.fit_transform(X=Xt, y=y)
  File "/home/anirban/sktime-fork/sktime/transformations/base.py", line 629, in fit_transform
    return self.fit(X, y).transform(X, y)
  File "/home/anirban/sktime-fork/sktime/transformations/base.py", line 449, in fit
    self._vectorize("fit", X=X_inner, y=y_inner)
  File "/home/anirban/sktime-fork/sktime/transformations/base.py", line 1240, in _vectorize
    transformers_ = X.vectorize_est(
  File "/home/anirban/sktime-fork/sktime/datatypes/_vectorize.py", line 623, in vectorize_est
    cols_right_order = df_long.loc[:, 1].unique()
  File "/home/anirban/conda-environments/sktime/lib/python3.10/site-packages/pandas/core/indexing.py", line 1097, in __getitem__
    return self._getitem_tuple(key)
  File "/home/anirban/conda-environments/sktime/lib/python3.10/site-packages/pandas/core/indexing.py", line 1280, in _getitem_tuple
    return self._getitem_lowerdim(tup)
  File "/home/anirban/conda-environments/sktime/lib/python3.10/site-packages/pandas/core/indexing.py", line 1000, in _getitem_lowerdim
    section = self._getitem_axis(key, axis=i)
  File "/home/anirban/conda-environments/sktime/lib/python3.10/site-packages/pandas/core/indexing.py", line 1343, in _getitem_axis
    return self._get_label(key, axis=axis)
  File "/home/anirban/conda-environments/sktime/lib/python3.10/site-packages/pandas/core/indexing.py", line 1293, in _get_label
    return self.obj.xs(label, axis=axis)
  File "/home/anirban/conda-environments/sktime/lib/python3.10/site-packages/pandas/core/generic.py", line 4082, in xs
    return self[key]
  File "/home/anirban/conda-environments/sktime/lib/python3.10/site-packages/pandas/core/frame.py", line 3761, in __getitem__
    indexer = self.columns.get_loc(key)
  File "/home/anirban/conda-environments/sktime/lib/python3.10/site-packages/pandas/core/indexes/range.py", line 347, in get_loc
    raise KeyError(key) from err
KeyError: 1

I was expecting no error with recursive strategy, as I thought predictions will be made one step at a time using seen y, unseen X and already forecasted y values.

But if it fails, I was expecting an immediate failure complaining that fh and window_length combination is not compatible with y shape or something along these line.

Is this a bug, or is it a planned error by design?

(Tagging @hazrulakmal as you've answered questions on make_reduction before.)

[fkiraly]

sounds like a bug - should we move to bugs?
For recursive reducers, the length of fh should not matter.