How to convert datatype of column names in the pipeline having OneHotEncoder

[VyomkeshVyas]

Hi all. I am trying to build a pipeline with linear regressor, lag features (as WindowSummarizer within make_reduction) and date time features (as one hot encoded) and I am getting following error while calling fit() :
TypeError: Feature names are only supported if all input features have string names, but your input has ['int', 'str'] as feature name / column name types. If you want feature names to be stored and validated, you must convert them all to strings, by using X.columns = X.columns.astype(str) for example. Otherwise you can remove feature / column names from your input data, or convert them all to a non-string data type.

Reproducible Example

from sktime.datasets import load_airline
from sktime.transformations.series.date import DateTimeFeatures
from sklearn.linear_model import LinearRegression
from sklearn.preprocessing import OneHotEncoder  
from sktime.transformations.series.adapt import TabularToSeriesAdaptor
from sktime.forecasting.compose import make_reduction
from sktime.transformations.series.summarize import WindowSummarizer
from sktime.forecasting.model_selection._split import temporal_train_test_split
import pandas as pd
import numpy as np

# Loading data
y = load_airline()

# train-test split
y_train, y_test = temporal_train_test_split(y)
X_train, X_test = pd.DataFrame(index = y_train.index), pd.DataFrame(index= y_test.index)

# Defining linear model
linear_mod = LinearRegression()

# Defining date time features
FEATURES = ['month_of_year', 'month_of_quarter']
dt_trafo = DateTimeFeatures(ts_freq = 'H', manual_selection=FEATURES, keep_original_columns=False)

# Defining lag features
kwargs = {
    "lag_feature": {
        'lag': [1,4,7]
    }
}
lg_trafo = WindowSummarizer(n_jobs=1, **kwargs)

# One-hot encoding date time features
enc = OneHotEncoder(sparse = False,drop='first')

# Reducing regressor to make it sktime compatible
forecaster = make_reduction(estimator= linear_mod, 
                            strategy="recursive",
                            window_length=None,
                            transformers=[lg_trafo],
                            pooling='global')

# Creating pipeline
pipe = (dt_trafo*TabularToSeriesAdaptor(enc))**forecaster

# Fitting training data
pipe.fit(y_train,X=X_train)

# Making predictions
pipe.predict(X=X_test, fh=np.arange(1, len(X_test) + 1))

Can someone help me in fixing this error?

[fkiraly]

Hm, this sounds like sklearn's input checks for pandas.DataFrame-s got more aggressive recently.

My guess is that something in the pipeline creates columns with inhomogenous types, and sklearn no longer likes that.

Let's move this to a bug report.

[fkiraly]

here now: #5165

[VyomkeshVyas]

Thanks for confirming. 👍