ruidongz / deep_matrix_factorization_models Goto Github PK

Do you know if there is any official code for the paper? If you have the link can you please share that?

def getTestNeg(self, testData, negNum):
    user = []
    item = []
    # testData是一个列表，列表元素是三元组(user, item, rate)
    for s in testData: # 对于测试集里每一个正例
        # tmp_user和tmp_item用来保存当前用户的测试集样本
        tmp_user = []
        tmp_item = []
        u = s[0]
        i = s[1]
        tmp_user.append(u)
        tmp_item.append(i)
        for t in range(negNum): # 都要生成negNum个负例
            j = np.random.randint(self.shape[1]) # shape[1]是物品数，这里即任取一件物品
            while (u, j) in self.trainDict: # 如果取到的user-item对是已在训练集里的，就重新取
                j = np.random.randint(self.shape[1])
            tmp_user.append(u)
            tmp_item.append(j)
        # user和item用来保存全部用户的测试集样本
        user.append(tmp_user)
        item.append(tmp_item)
    return [np.array(user), np.array(item)]

测试时，每个用户只取100个物品计算预测值，这里生成负例时只排除了出现在trainDict中的，所以如果随机生成的user-item对本身就是测试集中的正例，那么就会造成同一个user-item对在测试集中既为正例又为负例的情况，若采用HR和NDCG评估会使效果偏好。

另外，在getData时，应将离散的ID转换为连续编号，ML-1M数据集有3706部电影，但电影ID最小值是1，最大值是3952，不是连续的。

In DataSet.py, the func "getTrainTest" lost the last uesr and his ratings for train, if i understand correctly.

Problem:
when I conducts an experiment on Pinterest-20 dataset, the tensorflow give errors "ValueError: Cannot create a tensor proto whose content is larger than 2GB.".
Errors occur on the "self.user_item_embedding = tf.convert_to_tensor(self.dataSet.getEmbedding())".

Now, I solves it by using tf.variable.
Such as: https://blog.csdn.net/fjssharpsword/article/details/96431553

The change of code is :
def add_embedding_matrix(self): self.matrix_init = tf.placeholder(tf.float32, shape=(self.shape[0], self.shape[1])) matrix = tf.Variable(self.matrix_init) self.user_item_embedding = tf.convert_to_tensor(matrix) #self.user_item_embedding = tf.convert_to_tensor(self.dataSet.getEmbedding()) self.item_user_embedding = tf.transpose(self.user_item_embedding)

def init_sess(self): self.config = tf.ConfigProto() self.config.gpu_options.allow_growth = True self.config.allow_soft_placement = True self.sess = tf.Session(config=self.config) #self.sess.run(tf.global_variables_initializer()) self.sess.run(tf.global_variables_initializer(), feed_dict={self.matrix_init: self.dataSet.getEmbedding()})

Is it works? I don't know! Can you give me other effective solutions, thanks!

Problem:
When I sets the number of latent factors is 8 or 16, the experimental results of HR and NDCG all are 1.0. It is not reasonable.
parser.add_argument('-userLayer', action='store', dest='userLayer', default=[512, 8]) parser.add_argument('-itemLayer', action='store', dest='itemLayer', default=[1024, 8])

By debugging, I finds that the self.y_ always is 1e-6.
self.y_ = tf.reduce_sum(tf.multiply(user_out, item_out), axis=1, keepdims=False) / (norm_item_output* norm_user_output) self.y_ = tf.maximum(1e-6, self.y_)

In your paper, the experimental results showed that the HR and NDCG is normal when the number of latent factors is 8 or 16, can you give me the solution, thanks!

在我的电脑上跑15次迭代后HR和NDCG就=1，实验的结果也没有原文中那么好，想请问下原因

跑了下这个代码，效果和论文中单layer也就是 DMF-1-nce 差不多；

怎么修改参数去对照论文中的两层效果呢？