LightGBM/sklearn pipeline transformers are not run on fit_params['eval_set']

Question

When using GridSearchCV using early_stopping_rounds or an external test set in combination with a Pipeline it seems that eval_set are ignored by the pipeline. The fit function is only applied to the training data and the eval_set data is just passed to the final estimator without running the Transformers on them.

Is there a good way to work around this? I've attached a small example that shows that the eval_set is not transformed by the pipeline. I have read that it is possible to extend the classifier somehow but I am not sure how to access the pipeline object from it.

from sklearn.base import BaseEstimator
from sklearn.base import TransformerMixin
from sklearn.utils.validation import check_array, check_is_fitted
from sklearn.pipeline import Pipeline
from sklearn.model_selection import GridSearchCV
from sklearn.model_selection import train_test_split

import lightgbm as lgbm

import numpy as np

class Transformer(BaseEstimator, TransformerMixin):
    def __init__(self):
        pass

    def get_params(self, deep=True):
        return dict()

    def fit(self, X, y=None):
        X = check_array(X, dtype=object)
        print(X.shape)
        self.input_shape_ = X.shape
        return self

    def set_params(self, **parameters):
        self.__dict__.update(parameters)
        return self

    def transform(self, X):
        # Check is fit had been called
        check_is_fitted(self, ['input_shape_'])

        # Input validation
        X = check_array(X, dtype=object)
        Xt = np.zeros((len(X), 1), dtype=np.float32)

        for i in xrange(Xt.shape[0]):
            Xt[i] = np.float32(X[i][0].s)**2.0
        print(Xt)
        return Xt

class Foo:
    def __init__(self, s):
        self.s = s

if __name__ == '__main__':
    x = np.array([Foo(x) for x in xrange(10)]).reshape(-1, 1)
    y = np.array([0, 1, 0, 1, 0, 1, 0, 1, 0, 1])
    x_train, x_test, y_train, y_test = train_test_split(x, y, stratify=y, test_size=0.2, random_state=42)

    params = {'lgbm__learning_rate': [0.05, 0.1]}
    """
    static_params = {'n_estimators': 100,  # 0,
                     }
    """
    static_params = {'n_estimators': 100,  # 0,
                     'early_stopping_rounds': 5,
                     'eval_metric': 'binary_logloss',
                     'is_unbalance': False,
                     'eval_set': [[x_test, y_test]]
                     }



    pipe = Pipeline(steps=[('transformer', Transformer()), ('lgbm', lgbm.LGBMClassifier(**static_params))])
    estimator = GridSearchCV(pipe, scoring='roc_auc', param_grid=params, cv=2, n_jobs=-1)
    print(x_train)
    print(y_train)
    estimator.fit(x_train, y_train)