TunedThreasholdClassifierCV failing inside a SearchCV object

[adrinjalali]

I changed the existing example slightly, to put the estimator inside the SearchCV instead of tuning after the search. Here's the reproducer:

# %%
from sklearn.datasets import fetch_openml

# %%
credit_card = fetch_openml(data_id=1597, as_frame=True, parser="pandas")
credit_card.frame.info()

# %%
columns_to_drop = ["Class"]
data = credit_card.frame.drop(columns=columns_to_drop)
target = credit_card.frame["Class"].astype(int)

# %%
def business_metric(y_true, y_pred, amount):
    mask_true_positive = (y_true == 1) & (y_pred == 1)
    mask_true_negative = (y_true == 0) & (y_pred == 0)
    mask_false_positive = (y_true == 0) & (y_pred == 1)
    mask_false_negative = (y_true == 1) & (y_pred == 0)
    fraudulent_refuse = (mask_true_positive.sum() * 50) + amount[
        mask_true_positive
    ].sum()
    fraudulent_accept = -amount[mask_false_negative].sum()
    legitimate_refuse = mask_false_positive.sum() * -5
    legitimate_accept = (amount[mask_true_negative] * 0.02).sum()
    return fraudulent_refuse + fraudulent_accept + legitimate_refuse + legitimate_accept


# %%
import sklearn
from sklearn.metrics import make_scorer

sklearn.set_config(enable_metadata_routing=True)
business_scorer = make_scorer(business_metric).set_score_request(amount=True)

# %%
amount = credit_card.frame["Amount"].to_numpy()

# %%
from sklearn.model_selection import train_test_split

data_train, data_test, target_train, target_test, amount_train, amount_test = (
    train_test_split(
        data, target, amount, stratify=target, test_size=0.5, random_state=42
    )
)

# %%
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import GridSearchCV
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import TunedThresholdClassifierCV

logistic_regression = make_pipeline(StandardScaler(), LogisticRegression())

tuned_model = TunedThresholdClassifierCV(
    estimator=logistic_regression,
    scoring=business_scorer,
    thresholds=100,
    n_jobs=2,
)

param_grid = {"estimator__logisticregression__C": np.logspace(-6, 6, 13)}
model = GridSearchCV(tuned_model, param_grid, scoring=business_scorer).fit(
    data_train, target_train, amount=amount_train
)

It gives:

AttributeError: 'TunedThresholdClassifierCV' object has no attribute '_response_method'. Did you mean: 'response_method'?

This solves the issue temporarily:

tuned_model._response_method = "predict_proba"

cc @glemaitre @jeremiedbb

I'd argue this needs to be fixed for the release.

[glemaitre]

I open #29021 where I replaced setting a private _response_method in fit by a private method _get_response_method that can be called before fit to handle the metadata routing.