Keras makes it very easy to implement deep-learning models, however these are not compatible with scikit-learn out-of-the-box. This prohibits our models to be tested using sciki-learn's methods GridSearchCV and cross_val_score.
I created this MLP class to be compatible with scikit-learn that contains the fit, predict, and predict_proba methods.
Quick guide
Initialize your classifier
from mlp import MLP
clf = MLP(n_hidden=10, n_deep=3, l1_norm=0, drop=0.1, verbose=0)
Now evaluate your classifier with scikit-learn's cross_val_score
from sklearn.cross_validation import cross_val_score
scores = cross_val_score(clf, data, label, cv=5, n_jobs=-1, scoring='f1_weighted')
print(scores)
See a complete example in https://github.com/alvarouc/mlp/blob/master/examples/moon_sklearn.ipynb