In-sample predictions are not back transformed in ForecasterAutoreg()

[schoulten]

The problem

In order to generate in-sample predictions (aka fitted values), you need to create training matrices with .create_train_X_y() and use it with .predict() in the internal regressor, as described in the docs. But when any transformation is given to the ForecasterAutoreg(), it appears that the in-sample predictions are not being reverted to the original scale of the data.

Is it something that I missing or there is a way to revert the transformation?

Reproducible example

# Libraries
# ==============================================================================
import pandas as pd
import numpy as np
from skforecast.ForecasterAutoreg import ForecasterAutoreg
from skforecast.datasets import fetch_dataset
from sklearn.ensemble import RandomForestRegressor
from sklearn.preprocessing import PowerTransformer

# Download data
# ==============================================================================
data = fetch_dataset(name = "h2o_exog", raw = False)["y"]

# Split train-test
# ==============================================================================
steps = 36
data_train = data[:-steps]
data_test  = data[-steps:]

# Plot
# ==============================================================================
pd.concat([data_train.rename("train"), data_test.rename("test")], axis = "columns").plot()

# Create and fit forecaster without transformer
# ==============================================================================
forecaster_notrans = ForecasterAutoreg(
    regressor = RandomForestRegressor(random_state = 123),
    lags = 12
    )

forecaster_notrans.fit(y = data_train)

forecaster_notrans

# Create training matrices
# ==============================================================================
X_train, y_train = forecaster_notrans.create_train_X_y(data_train)
X_train.head()

# Predict using the internal regressor
# ==============================================================================
predictions_1 = forecaster_notrans.regressor.predict(X_train)
predictions_1[:4]

# Plot predictions
# ==============================================================================
pd.concat([
    pd.Series(predictions_1, name = "fitted", index = data_train.index[forecaster_notrans.max_lag:]),
    data_train.rename("train")
], axis = "columns").plot(title = "No transformer");

# Create and fit forecaster with transformer
# ==============================================================================
forecaster_trans = ForecasterAutoreg(
    regressor = RandomForestRegressor(random_state = 123),
    lags = 12,
    transformer_y = PowerTransformer()
    )

forecaster_trans.fit(y = data_train)

forecaster_trans

# Create training matrices
# ==============================================================================
X_train, y_train = forecaster_trans.create_train_X_y(data_train)
X_train.head()

# Predict using the internal regressor
# ==============================================================================
predictions_1 = forecaster_trans.regressor.predict(X_train)
predictions_1[:4]

# Plot predictions
# ==============================================================================
pd.concat([
    pd.Series(predictions_1, name = "fitted", index = data_train.index[forecaster_trans.max_lag:]),
    data_train.rename("train")
], axis = "columns").plot(title = "With transformer");

# Out of sample predictions are OK
# ==============================================================================
predictions_3 = forecaster_trans.predict(steps = steps)
predictions_3.head(3)

# Plot predictions
# ==============================================================================
pd.concat([
    pd.Series(predictions_1, name = "fitted", index = data_train.index[forecaster_trans.max_lag:]),
    data_train.rename("train"),
    data_test.rename("test"),
    predictions_3.rename("forecast")
], axis = "columns").plot(title = "With transformer");

Session information

Preparing metadata (setup.py) ... done
  Building wheel for session-info (setup.py) ... done
Click to view session information
-----
matplotlib          3.7.1
numpy               1.25.2
pandas              2.0.3
session_info        1.0.0
skforecast          0.12.1
sklearn             1.2.2
-----
Click to view modules imported as dependencies
PIL                 9.4.0
backcall            0.2.0
certifi             2024.06.02
cffi                1.16.0
cloudpickle         2.2.1
cycler              0.12.1
cython_runtime      NA
dateutil            2.8.2
debugpy             1.6.6
decorator           4.4.2
defusedxml          0.7.1
google              NA
httplib2            0.22.0
ipykernel           5.5.6
ipython_genutils    0.2.0
joblib              1.4.2
kiwisolver          1.4.5
matplotlib_inline   0.1.7
mpl_toolkits        NA
numexpr             2.10.0
packaging           24.1
pexpect             4.9.0
pickleshare         0.7.5
pkg_resources       NA
platformdirs        4.2.2
portpicker          NA
prompt_toolkit      3.0.47
psutil              5.9.5
ptyprocess          0.7.0
pyarrow             14.0.2
pydev_ipython       NA
pydevconsole        NA
pydevd              2.9.5
pydevd_file_utils   NA
pydevd_plugins      NA
pydevd_tracing      NA
pygments            2.16.1
pyparsing           3.1.2
pytz                2023.4
scipy               1.11.4
setuptools          67.7.2
sitecustomize       NA
six                 1.16.0
socks               1.7.1
sphinxcontrib       NA
storemagic          NA
threadpoolctl       3.5.0
tornado             6.3.3
traitlets           5.7.1
typing_extensions   NA
wcwidth             0.2.13
zmq                 24.0.1
zoneinfo            NA
-----
IPython             7.34.0
jupyter_client      6.1.12
jupyter_core        5.7.2
notebook            6.5.5
-----
Python 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0]
Linux-6.1.85+-x86_64-with-glibc2.35
-----
Session information updated at 2024-06-14 11:16