ValueError when using [GlobalSklearnTransformer+PowerTransformer]

[wasf84]

What happened + What you expected to happen

Hello! o/

I was reading the manual to learn about new ways to transform data. However, when I attempt to apply the 'GlobalSklearnTransformer' with a 'PowerTransformer', I consistently encounter an error. All I did was try to apply the example shown in the manual here https://nixtlaverse.nixtla.io/mlforecast/target_transforms.html to my data, but the error persists. This example from the manual is exactly what I want to accomplish with my data—it would be perfect for my use case.

simple_train.xlsx

---

ValueError Traceback (most recent call last)
Cell In[184], line 1
----> 1 df_cv = mlf.cross_validation(
2 df = simple_train[["ds", "unique_id", "y"]],
3 n_windows = 5,
4 h = 15,
5 step_size = 15,
6 refit = True,
7 static_features = [],
8 )

File ~/bin/miniconda3/envs/py311/lib/python3.11/site-packages/mlforecast/forecast.py:958, in MLForecast.cross_validation(self, df, n_windows, h, id_col, time_col, target_col, step_size, static_features, dropna, keep_last_n, refit, max_horizon, before_predict_callback, after_predict_callback, prediction_intervals, level, input_size, fitted, as_numpy, weight_col)
956 else:
957 X_df = None
--> 958 y_pred = self.predict(
959 h=h,
960 before_predict_callback=before_predict_callback,
961 after_predict_callback=after_predict_callback,
962 new_df=train if not should_fit else None,
963 level=level,
964 X_df=X_df,
965 )
966 y_pred = ufp.join(y_pred, cutoffs, on=id_col, how="left")
967 result = ufp.join(
968 valid[[id_col, time_col, target_col]],
969 y_pred,
970 on=[id_col, time_col],
971 )

File ~/bin/miniconda3/envs/py311/lib/python3.11/site-packages/mlforecast/forecast.py:738, in MLForecast.predict(self, h, before_predict_callback, after_predict_callback, new_df, level, X_df, ids)
735 else:
736 ts = self.ts
--> 738 forecasts = ts.predict(
739 models=self.models_,
740 horizon=h,
741 before_predict_callback=before_predict_callback,
742 after_predict_callback=after_predict_callback,
743 X_df=X_df,
744 ids=ids,
745 )
746 if level is not None:
747 if self._cs_df is None:

File ~/bin/miniconda3/envs/py311/lib/python3.11/site-packages/mlforecast/core.py:856, in TimeSeries.predict(self, models, horizon, before_predict_callback, after_predict_callback, X_df, ids)
854 preds = ufp.assign_columns(preds, col, ga.data)
855 else:
--> 856 preds = tfm.inverse_transform(preds)
857 return preds

File ~/bin/miniconda3/envs/py311/lib/python3.11/site-packages/mlforecast/target_transforms.py:333, in GlobalSklearnTransformer.inverse_transform(self, df)
329 df = ufp.copy_if_pandas(df, deep=False)
330 cols_to_transform = [
331 c for c in df.columns if c not in (self.id_col, self.time_col)
332 ]
--> 333 transformed = self.transformer_.inverse_transform(
334 df[cols_to_transform].to_numpy()
335 )
336 return ufp.assign_columns(df, cols_to_transform, transformed)

File ~/bin/miniconda3/envs/py311/lib/python3.11/site-packages/sklearn/preprocessing/_data.py:3379, in PowerTransformer.inverse_transform(self, X)
3348 """Apply the inverse power transformation using the fitted lambdas.
3349
3350 The inverse of the Box-Cox transformation is given by::
(...)
3376 The original data.
3377 """
3378 check_is_fitted(self)
-> 3379 X = self._check_input(X, in_fit=False, check_shape=True)
3381 if self.standardize:
3382 X = self._scaler.inverse_transform(X)

File ~/bin/miniconda3/envs/py311/lib/python3.11/site-packages/sklearn/preprocessing/_data.py:3513, in PowerTransformer._check_input(self, X, in_fit, check_positive, check_shape)
3495 def check_input(self, X, in_fit, check_positive=False, check_shape=False):
3496 """Validate the input before fit and transform.
3497
3498 Parameters
(...)
3511 If True, check that n_features matches the length of self.lambdas
3512 """
-> 3513 X = self._validate_data(
3514 X,
3515 ensure_2d=True,
3516 dtype=FLOAT_DTYPES,
3517 force_writeable=True,
3518 copy=self.copy,
3519 force_all_finite="allow-nan",
3520 reset=in_fit,
3521 )
3523 with warnings.catch_warnings():
3524 warnings.filterwarnings("ignore", r"All-NaN (slice|axis) encountered")

File ~/bin/miniconda3/envs/py311/lib/python3.11/site-packages/sklearn/base.py:654, in BaseEstimator._validate_data(self, X, y, reset, validate_separately, cast_to_ndarray, **check_params)
651 out = X, y
653 if not no_val_X and check_params.get("ensure_2d", True):
--> 654 self._check_n_features(X, reset=reset)
656 return out

File ~/bin/miniconda3/envs/py311/lib/python3.11/site-packages/sklearn/base.py:443, in BaseEstimator.check_n_features(self, X, reset)
440 return
442 if n_features != self.n_features_in:
--> 443 raise ValueError(
444 f"X has {n_features} features, but {self.class.name} "
445 f"is expecting {self.n_features_in_} features as input."
446 )

ValueError: X has 6 features, but PowerTransformer is expecting 1 features as input.

Versions / Dependencies

Python
3.11.11

Zorin OS
17.2

mlforecast
1.0.1

sklearn
1.5.2

pandas
2.2.3

numpy
1.26.4

scipy
1.14.0

Reproduction script

from sklearn.preprocessing import PowerTransformer
from mlforecast.target_transforms import GlobalSklearnTransformer

sk_yeojohnson = PowerTransformer(method='yeo-johnson', standardize=False)
yeojohnson_global = GlobalSklearnTransformer(sk_yeojohnson)

mdl_lr = LinearRegression()
mdl_cb = CatBoostRegressor(loss_function="MAE", verbose=False, allow_writing_files=False, has_time=True)
mdl_xgb = XGBRegressor(verbosity=0)
mdl_lgbm = LGBMRegressor(objective="regression_l1", verbosity    = -1)
mdl_rf = RandomForestRegressor(verbose=0, criterion="friedman_mse")
mdl_gb = GradientBoostingRegressor(verbose=0, loss="absolute_error")

models_ml = [mdl_lr, mdl_cb, mdl_xgb, mdl_lgbm, mdl_rf, mdl_gb]

mlf = MLForecast(models=models_ml, freq='D', lags=list(range(1, 31)), target_transforms=[yeojohnson_global])

df_cv = mlf.cross_validation(df=simple_train[["ds", "unique_id", "y"]], n_windows=5, h=15, step_size=15, refit=True, static_features = [])

Issue Severity

None

[jmoralez]

Hey. Thanks for the report, a fix will be released soon.