keras自定义目标函数(二)

from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.model_selection import train_test_split
from tensorflow.keras.layers import Dense ,Concatenate
#import keras
import keras.backend as K

from tensorflow import keras
from tensorflow.keras.layers import concatenate
from tensorflow.keras import Sequential,Model
from tensorflow.keras.layers import Dense ,Concatenate,Input,BatchNormalization

dense_f=78
hero_f=26
giftbag_f=227
recent_f=1627

def build_model(dense_f,hero_f , giftbag_f, recent_f):
    
    dense_inputs = Input(shape=(dense_f, ))
    hero_inputs = Input(shape=(hero_f, ))
    gift_inputs = Input(shape=(giftbag_f, ))
    recent_inputs = Input(shape=(recent_f, ))
    
    gift_dense = Dense(256, activation='relu')(gift_inputs)
#    gift_dense= keras.layers.BatchNormalization()(gift_dense)
    
    
    recent_dense = Dense(256, activation='relu')(recent_inputs)
    recent_dense=BatchNormalization()(recent_dense)
    
    merge_inputs = concatenate([dense_inputs ,hero_inputs,gift_dense,recent_dense],axis=1)


    flow_dense = Dense(256, activation='relu')(merge_inputs)
    flow_dense = BatchNormalization()( flow_dense)
    flow_dense = Dense(128, activation='relu')(flow_dense)
    flow_dense = BatchNormalization()( flow_dense)
    flow_dense = Dense(64, activation='relu')(flow_dense)
    flow_dense = BatchNormalization()( flow_dense)
    outputs = Dense(n_classes+1, activation='softmax')(flow_dense)
    
#    b = K.slice(outputs, [0, 0], [-1,128])
#    
#    b1 = K.slice(outputs, [-1, 128], [-1,1])
#    
    def mycrossentropy(y_true, y_pred, e=0.3):
        b=y_pred[:,:128]
        b1=y_pred[:,128:]

        c=y_true[:,:128]
        c1=y_true[:,128:129]
        loss1 = K.categorical_crossentropy(b, c)
        loss2 =  K.mean(K.square(b1 -c1), axis=-1)
        return   (1-e)*loss1 * e*loss2

    model = keras.Model(inputs=[dense_inputs, hero_inputs, gift_inputs, recent_inputs], outputs=outputs)
    opt = keras.optimizers.Adam()
    model.compile(optimizer=opt, loss=mycrossentropy)
    return model

model = build_model(dfdense.shape[1], hero_zo.shape[1], gift_tensor.shape[1], recentt1.shape[1])


model.fit([dfdense,  hero_zo,  gift_tensor, recentt1],yyp, 
              
              epochs=100, verbose=1, batch_size=64)