Could you please let me know how to set class-weight for imbalanced classes in KerasClassifier while it is used inside the GridSearchCV?

# Use scikit-learn to grid search the batch size and epochs
from collections import Counter
from sklearn.model_selection import train_test_split,StratifiedKFold,learning_curve,validation_curve,GridSearchCV
from sklearn.datasets import make_classification
from sklearn.preprocessing import StandardScaler
import numpy as np
from sklearn.model_selection import GridSearchCV
from keras.models import Sequential
from keras.layers import Dense
from keras.wrappers.scikit_learn import KerasClassifier
from sklearn.metrics import classification_report
import pandas as pd
from sklearn.pipeline import Pipeline

# Function to create model, required for KerasClassifier
def create_model():
    # create model
    model = Sequential()
    model.add(Dense(12, input_dim=20, activation='relu'))
    model.add(Dense(1, activation='sigmoid'))
    # Compile model
    model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
    return model
# fix random seed for reproducibility
seed = 7
# load dataset
X, y = make_classification(n_classes=2, class_sep=2,weights=[0.95, 0.05], n_informative=3, n_redundant=2, flip_y=0, n_features=20, n_clusters_per_class=1, n_samples=1000, random_state=10)
print('Original dataset shape {}'.format(Counter(y)))

ln = X.shape

X_train, X_test, y_train, y_test = train_test_split(X, y,random_state=0)

# create model
model = KerasClassifier(build_fn=create_model, verbose=0)
pipeline = Pipeline(steps=[('scaler', st),
                               ('clf', model )])
# define the grid search parameters
batch_size = [20, 40, 60, 80, 100]
epochs = [ 50, 100]
param_grid = dict(clf__batch_size=batch_size, clf__epochs=epochs)
cv = StratifiedKFold(n_splits=5, random_state=42)

grid = GridSearchCV(estimator=pipeline, param_grid=param_grid,cv=cv,scoring="f1")
grid_result = grid.fit(X_train, y_train)
# summarize results
print("Best: %f using %s" % (grid_result.best_score_, grid_result.best_params_))
# Predictions
ypred = grid_result.predict(X_train)
print(classification_report(y_train, ypred))
ypred2 = grid_result.predict(X_test)
print(classification_report(y_test, ypred2))
  • $\begingroup$ sorry for late response. I guess you already have found the answer. $\endgroup$ – Media May 26 '18 at 15:34
  • $\begingroup$ Actually it depends on different aspects. What are your data like? $\endgroup$ – Media Jun 17 '18 at 15:25
grid_result = grid.fit(X_train, y_train, clf__class_weight={0:0.95, 1:0.05})

FYI, per the docs fit_params should no longer be passed to the GridSearchCV constructor as a dict, but should be passed directly to fit as above.


  • $\begingroup$ Thanks a lot but it seems it should be changed into: clf__class_weight={0:0.05,1:0.95}. Therefore, it is not possible to tune class_weight in a way that is done for svc or logistic regression. $\endgroup$ – ebrahimi May 24 '18 at 5:33
  • $\begingroup$ You're right and I've made the change. I don't believe you can tune fit_params with GridSearchCV $\endgroup$ – Bert Kellerman May 24 '18 at 12:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.