keras中自定义验证集的性能评估（ROC,AUC）

在keras中自带的性能评估有准确性以及loss，当需要以auc作为评价验证集的好坏时，就得自己写个评价函数了：

from sklearn.metrics import roc_auc_score

from keras import backend as K

# AUC for a binary classifier
def auc(y_true, y_pred):
    ptas = tf.stack([binary_PTA(y_true,y_pred,k) for k in np.linspace(0, 1, 1000)],axis=0)
    pfas = tf.stack([binary_PFA(y_true,y_pred,k) for k in np.linspace(0, 1, 1000)],axis=0)
    pfas = tf.concat([tf.ones((1,)) ,pfas],axis=0)
    binSizes = -(pfas[1:]-pfas[:-1])
    s = ptas*binSizes
    return K.sum(s, axis=0)
#-----------------------------------------------------------------------------------------------------------------------------------------------------
# PFA, prob false alert for binary classifier
def binary_PFA(y_true, y_pred, threshold=K.variable(value=0.5)):
    y_pred = K.cast(y_pred >= threshold, 'float32')
    # N = total number of negative labels
    N = K.sum(1 - y_true)
    # FP = total number of false alerts, alerts from the negative class labels
    FP = K.sum(y_pred - y_pred * y_true)
    return FP/N
#-----------------------------------------------------------------------------------------------------------------------------------------------------
# P_TA prob true alerts for binary classifier
def binary_PTA(y_true, y_pred, threshold=K.variable(value=0.5)):
    y_pred = K.cast(y_pred >= threshold, 'float32')
    # P = total number of positive labels
    P = K.sum(y_true)
    # TP = total number of correct alerts, alerts from the positive class labels
    TP = K.sum(y_pred * y_true)
    return TP/P

#接着在模型的compile中设置metrics
#如下例子，我用的是RNN做分类

from keras.models import Sequential
from keras.layers import Dense, Dropout
import keras
from keras.layers import GRU

model = Sequential()
model.add(keras.layers.core.Masking(mask_value=0., input_shape=(max_lenth, max_features))) #masking用于变长序列输入
model.add(GRU(units=n_hidden_units,activation='selu',kernel_initializer='orthogonal', recurrent_initializer='orthogonal',
              bias_initializer='zeros', kernel_regularizer=regularizers.l2(0.01), recurrent_regularizer=regularizers.l2(0.01),
              bias_regularizer=None, activity_regularizer=None, kernel_constraint=None, recurrent_constraint=None,
              bias_constraint=None, dropout=0.5, recurrent_dropout=0.0, implementation=1, return_sequences=False,
              return_state=False, go_backwards=False, stateful=False, unroll=False))   
model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))

model.compile(loss='binary_crossentropy',
              optimizer='adam',
              metrics=[auc])  #写入自定义评价函数

接下来就自己作预测了...

方法二：

from sklearn.metrics import roc_auc_score
import keras
class RocAucMetricCallback(keras.callbacks.Callback):
    def __init__(self, predict_batch_size=1024, include_on_batch=False):
        super(RocAucMetricCallback, self).__init__()
        self.predict_batch_size=predict_batch_size
        self.include_on_batch=include_on_batch

    def on_batch_begin(self, batch, logs={}):
        pass

    def on_batch_end(self, batch, logs={}):
        if(self.include_on_batch):
            logs['roc_auc_val']=float('-inf')
            if(self.validation_data):
                logs['roc_auc_val']=roc_auc_score(self.validation_data[1], 
                                                  self.model.predict(self.validation_data[0],
                                                                     batch_size=self.predict_batch_size))

    def on_train_begin(self, logs={}):
        if not ('roc_auc_val' in self.params['metrics']):
            self.params['metrics'].append('roc_auc_val')

    def on_train_end(self, logs={}):
        pass

    def on_epoch_begin(self, epoch, logs={}):
        pass

    def on_epoch_end(self, epoch, logs={}):
        logs['roc_auc_val']=float('-inf')
        if(self.validation_data):
            logs['roc_auc_val']=roc_auc_score(self.validation_data[1], 
                                              self.model.predict(self.validation_data[0],
                                                                 batch_size=self.predict_batch_size))


import numpy as np
import tensorflow as tf
from keras.models import Sequential
from keras.layers import Dense, Dropout
from keras.layers import GRU
import keras
from keras.callbacks import EarlyStopping
from sklearn.metrics import roc_auc_score
from keras import metrics

cb = [
    my_callbacks.RocAucMetricCallback(), # include it before EarlyStopping!
    EarlyStopping(monitor='roc_auc_val',patience=300, verbose=2,mode='max')
]
model = Sequential()
model.add(keras.layers.core.Masking(mask_value=0., input_shape=(max_lenth, max_features)))
# model.add(Embedding(input_dim=max_features+1, output_dim=64,mask_zero=True))
model.add(GRU(units=n_hidden_units,activation='selu',kernel_initializer='orthogonal', recurrent_initializer='orthogonal',
              bias_initializer='zeros', kernel_regularizer=regularizers.l2(0.01), recurrent_regularizer=regularizers.l2(0.01),
              bias_regularizer=None, activity_regularizer=None, kernel_constraint=None, recurrent_constraint=None,
              bias_constraint=None, dropout=0.5, recurrent_dropout=0.0, implementation=1, return_sequences=False,
              return_state=False, go_backwards=False, stateful=False, unroll=False))   #input_shape=(max_lenth, max_features),
model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))

model.compile(loss='binary_crossentropy',
              optimizer='adam',
              metrics=[auc]) #这里就可以写其他评估标准
model.fit(x_train, y_train, batch_size=train_batch_size, epochs=training_iters, verbose=2,
          callbacks=cb,validation_split=0.2,
          shuffle=True, class_weight=None, sample_weight=None, initial_epoch=0)

亲测有效