[Python] Using Ordinal Encoding with Pipeline when there is new data in test set

Trying to use pipeline of One-Hot/Ordinal encoding with different models using cross_val_score. When there is unseen data in test set I handle it like this for OHE:

ohe = make_column_transformer((OneHotEncoder(handle_unknown='ignore'), cate_cols),
remainder='passthrough')


however for ordinal encoder I cannot figure out how... I've tried:

oe = make_column_transformer((OrdinalEncoder(handle_unknown='use_encoded_value', unknown_value=np.nan), cate_cols),
remainder='passthrough')


when I run:

dt_reg = DecisionTreeRegressor()
model = ("DT_reg_oe", make_pipeline(oe, dt_reg))
cv_results = cross_val_score(model, X, y, cv=5, scoring=scoring)


this outputs: ValueError: Input contains NaN, infinity or a value too large for dtype('float32') which does not make sense because I've selected only columns that have dtypes category and object. I need to change category dtype to float(isn't this what ordinal encoding is supposed to do)? how can I use ordinal encoding for categorical columns inside pipeline?

Continue reading...