Convert Neural network to Keras Classifier

Question

I am training a Neural Network for Multi-Class classification. After succesfully training it and validating the model through cross validation, I would like to use this network inside a voting Classifier. In order to perform cross validation on my trained network I convert it to a Keras Classifier and then calculate its validation score. However when I parse the same exact "Keras Classifier" in the Voting Classifier method, I get the following error:

ValueError: The estimator KerasClassifier should be a classifier

The code can be seen below:

import random 
random.seed(42)

from keras.layers import Dense
from keras.models import Sequential
from keras.wrappers.scikit_learn import KerasClassifier
import tensorflow as tf
from sklearn.ensemble import VotingClassifier

def NeuralNetwork():
    model = Sequential()
    # define first hidden layer and visible layer
    model.add(Dense(600, input_dim=k, activation='relu'))
    model.add(Dense(20, activation='relu'))
    # define output layer
    model.add(Dense(3, activation='softmax'))
    # define loss and optimizer
    model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=["accuracy"])
    return model

NN = KerasClassifier(build_fn=NeuralNetwork , epochs=100, batch_size=100, verbose=0)
cross_val_score(NN, X_new_train,y_train_no_id, cv=3)


votingC = VotingClassifier(estimators=[ ('LR1', LR1),('LR2', LR2), ('XGB',XGB),('NN',NN)], voting='hard', n_jobs=4)   
#LR1 and LR2 are some other Logistic Regression estimators defined in another section
votingC = votingC.fit(X_new_train, y_train_no_id)
votingC.predict(X_new_test)

Answer 1

There are a couple of Github issues on this - see here and here. In short, while scikit-learn native models include an _estimator_type attribute:

from sklearn.linear_model import LogisticRegression
clf = LogisticRegression()
clf._estimator_type
# 'classifier'

this is not the case with a KerasClassifier; using your own NN gives

NN = KerasClassifier(build_fn=NeuralNetwork , epochs=100, batch_size=100, verbose=0)
NN._estimator_type

---------------------------------------------------------------------------

AttributeError                            Traceback (most recent call last)

<ipython-input-4-a2fd193c154f> in <module>()
----> 1 NN._estimator_type

AttributeError: 'KerasClassifier' object has no attribute '_estimator_type'

As reported in a comment here, setting the attribute manually seems to work:

NN._estimator_type = "classifier"

so add this line immediately after your NN definition.