keras用auc做metrics以及早停

import tensorflow as tf
from sklearn.metrics import roc_auc_score

def auroc(y_true, y_pred):
    return tf.py_func(roc_auc_score, (y_true, y_pred), tf.double)

# Build Model...

model.compile(loss='categorical_crossentropy', optimizer='adam',metrics=['accuracy', auroc])

完整例子:

def auc(y_true, y_pred):
    auc = tf.metrics.auc(y_true, y_pred)[1]
    K.get_session().run(tf.local_variables_initializer())
    return auc


def create_model_nn(in_dim,layer_size=200):
    model = Sequential()
    model.add(Dense(layer_size,input_dim=in_dim, kernel_initializer='normal'))
    model.add(BatchNormalization())
    model.add(Activation('relu'))
    model.add(Dropout(0.3))
    for i in range(2):
        model.add(Dense(layer_size))
        model.add(BatchNormalization())
        model.add(Activation('relu'))
        model.add(Dropout(0.3))
    model.add(Dense(1, activation='sigmoid'))
    adam = optimizers.Adam(lr=0.01)
    model.compile(optimizer=adam,loss='binary_crossentropy',metrics = [auc])    
    return model
####cv train
folds = StratifiedKFold(n_splits=5, shuffle=False, random_state=15)
oof = np.zeros(len(df_train))
predictions = np.zeros(len(df_test))
for fold_, (trn_idx, val_idx) in enumerate(folds.split(df_train.values, target2.values)):
    print("fold n°{}".format(fold_))
    X_train = df_train.iloc[trn_idx][features]
    y_train = target2.iloc[trn_idx]
    X_valid = df_train.iloc[val_idx][features]
    y_valid = target2.iloc[val_idx]
    model_nn = create_model_nn(X_train.shape[1])
    callback = EarlyStopping(monitor="val_auc", patience=50, verbose=0, mode='max')
    history = model_nn.fit(X_train, y_train, validation_data = (X_valid ,y_valid),epochs=1000,batch_size=64,verbose=0,callbacks=[callback])
    print('\n Validation Max score : {}'.format(np.max(history.history['val_auc'])))
    predictions += model_nn.predict(df_test[features]).ravel()/folds.n_splits

参考:
https://stackoverflow.com/questions/41032551/how-to-compute-receiving-operating-characteristic-roc-and-auc-in-keras

https://www.kaggle.com/c/santander-customer-transaction-prediction/discussion/80807#latest-482777

你可能感兴趣的:(ML,DL)