深度推荐模型——AFM [IJCAI 17][浙大、新加坡国立大学]

import tensorflow as tf
from tensorflow import keras
from utils import *

EPOCH = 10
BATCH_SIZE = 32
VEC_DIM = 10
DNN_LAYERS = [64, 128, 64]
DROPOUT_RATE = 0.5

base, test = loadData()
# 所有的特征各个类别值个数之和
FEAT_CATE_NUM = base.shape[1] - 1
K = tf.keras.backend


class PairWiseInteractionAttentionLayer(keras.layers.Layer):
    def __init__(self, vec_dim, attention_factor, **kwargs):
        self.vec_dim = vec_dim
        self.attention_factor = attention_factor
        super(PairWiseInteractionAttentionLayer, self).__init__(**kwargs)

    def build(self, input_shape):
        self.cnum = input_shape[1]
        self.cross_num = self.cnum * (self.cnum - 1) // 2
        self.W0 = self.add_weight(name='W0', shape=(self.vec_dim, self.attention_factor),
                                  initializer='uniform',trainable=True)
        self.W1 = self.add_weight(name='W1', shape=(self.attention_factor, 1), initializer='uniform',
                                  trainable=True)
        super(PairWiseInteractionAttentionLayer, self).build(input_shape)

    def call(self, Input, **kwargs):
        pi_emb = []
        for i in range(self.cnum):
            for j in range(i + 1, self.cnum):
                pi_emb.append(tf.multiply(Input[:, i, :], Input[:, j, :]))
        pi_emb = tf.stack(pi_emb, axis=1)  # [-1,coss_num,vec_dim]
        att = tf.matmul(tf.nn.relu(tf.matmul(pi_emb, self.W0)), self.W1)  # [-1,coss_num,1]
        att_score = tf.reshape(tf.nn.softmax(tf.reshape(att, shape=(-1, self.cross_num))),
                               shape=(-1, self.cross_num, 1))  # [-1,coss_num]
        weight_sum = tf.multiply(pi_emb, att_score)  # [-1,coss_num,vec_dim]
        weight_sum = tf.reduce_sum(weight_sum, axis=1)
        return weight_sum


def run():
    # 将所有的特征的各个类别值统一id化。x中每行为各特征的类别值的id
    val_x, val_y = getAllData(test)
    train_x, train_y = getAllData(base)

    model = keras.models.Sequential()
    model.add(keras.layers.Embedding(FEAT_CATE_NUM, VEC_DIM, input_length=val_x[0].shape[0]))
    model.add(PairWiseInteractionAttentionLayer(vec_dim=VEC_DIM, attention_factor=VEC_DIM))
    model.add(keras.layers.Dropout(rate=DROPOUT_RATE))
    for units in DNN_LAYERS:
        model.add(keras.layers.Dense(units, activation='relu'))
        model.add(keras.layers.Dropout(DROPOUT_RATE))
    model.add(keras.layers.Dense(1, activation='sigmoid'))

    model.compile(loss='binary_crossentropy', optimizer=tf.train.AdamOptimizer(0.001), metrics=[keras.metrics.AUC()])
    tbCallBack = keras.callbacks.TensorBoard(log_dir='./logs',
                                             histogram_freq=0,
                                             write_graph=True,
                                             write_grads=True,
                                             write_images=True,
                                             embeddings_freq=0,
                                             embeddings_layer_names=None,
                                             embeddings_metadata=None)

    model.fit(train_x, train_y, batch_size=BATCH_SIZE, epochs=EPOCH, verbose=2, validation_data=(val_x, val_y),
              callbacks=[tbCallBack])


run()