fc1 = PReLU()(fc1)
fc1 = BatchNormalization()(fc1)
fc1 = Dropout(0.2)(fc1)
outputs = Dense(1,activation='sigmoid')(fc1)
model = Model(input=inputs, output=outputs)
model.compile(
optimizer='adam',
loss = 'binary_crossentropy',
)
return model
class NN(BaseEstimator):
def __init__(self,input_shape,hidden=128):
self.input_shape = input_shape
self.hidden = hidden
def fit(self,X,y):
model = build_model(maxlen=self.input_shape,hidden=self.hidden)
model.fit(X,y,nb_epoch=250,batch_size=128,shuffle=True,verbose=0)
self.model=model
def predict_proba(self,X):
y_p = self.model.predict(X,batch_size=1024).ravel()
y_p_inv= np.ones(y_p.shape[0])-y_p
y_p = np.hstack([y_p_inv.reshape(-1,1),y_p.reshape(-1,1)])
return y_p
clf = NN(X.shape[1],128)
make_mf_classification(X ,y, clf, X_t, n_folds=5,seed=seed,nb_epoch=1,max_features=1.0,name='nn_2',path=path)
class NN(BaseEstimator):
def __init__(self, input_shape, hidden=128):
self.input_shape = input_shape
self.hidden = hidden
def fit(self, X, y):
model = build_model(maxlen=self.input_shape, hidden=self.hidden)
model.fit(X, y, epochs=250, batch_size=128, shuffle=True, verbose=1)
self.model = model
def predict_proba(self, X):
y_p = self.model.predict(X, batch_size=1024).ravel()
y_p_inv = np.ones(y_p.shape[0]) - y_p
y_p = np.hstack([y_p_inv.reshape(-1, 1), y_p.reshape(-1, 1)])
return y_p
scaler = MinMaxScaler()
scaler.fit(X)
clf = NN(X.shape[1], 128)
make_mf_classification(scaler.transform(X),
y,
clf,
scaler.transform(X_t),
n_folds=5,
seed=seed,
nb_epoch=1,
max_features=1.0,
name='nn_3',
path=path)