QRec: A Python Framework for quick implementation of recommender systems (TensorFlow Based)

你好，在Macbook上运行main.py出现了一个问题，下载了依赖mkl，还是出现这个：

/usr/local/bin/python2.7 /Users/feanlau/Desktop/RecQ/main/main.py
Traceback (most recent call last):
  File "/Users/feanlau/Desktop/RecQ/main/main.py", line 3, in <module>
    from RecQ import RecQ
  File "/Users/feanlau/Desktop/RecQ/main/RecQ.py", line 9, in <module>
    import mkl
ImportError: No module named mkl

Process finished with exit code 1

请问一下，这个问题如何解决

When running the RSGAN algorithm, this error below appears frequently that I can hardly get the predicting process completed. I was wondering whether the "Process Process-2" is right or not. Is it the problem of multi-process?

Predicting [1]...
RSGAN [1] progress:0/1753
RSGAN [1] progress:100/1753
RSGAN [1] progress:200/1753
RSGAN [1] progress:300/1753
RSGAN [1] progress:400/1753
Process Process-2:
Traceback (most recent call last):
  File "/home/azure/miniconda3/envs/py2/lib/python2.7/multiprocessing/process.py", line 267, in _bootstrap
    self.run()
  File "/home/azure/miniconda3/envs/py2/lib/python2.7/multiprocessing/process.py", line 114, in run
    self._target(*self._args, **self._kwargs)
  File "/home/azure/RecQ-master/main/RecQ.py", line 116, in run
    measure[order] = algor.execute()
  File "../baseclass/Recommender.py", line 257, in execute
    self.evalRanking()
  File "../baseclass/Recommender.py", line 153, in evalRanking
    predictedItems = self.predictForRanking(user)
  File "../algorithm/ranking/RSGAN.py", line 448, in predictForRanking
    res = self.sess.run(self.d_output, {self.u_idx:[u]})
  File "/home/azure/miniconda3/envs/py2/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 950, in run
    run_metadata_ptr)
  File "/home/azure/miniconda3/envs/py2/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1173, in _run
    feed_dict_tensor, options, run_metadata)
  File "/home/azure/miniconda3/envs/py2/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1350, in _do_run
    run_metadata)
  File "/home/azure/miniconda3/envs/py2/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1370, in _do_call
    raise type(e)(node_def, op, message)
InvalidArgumentError: indices[0] = 1875 is not in [0, 1795)
         [[node discriminator/u_e (defined at ../algorithm/ranking/RSGAN.py:334) ]]

Errors may have originated from an input operation.
Input Source operations connected to node discriminator/u_e:
 U/read (defined at ../baseclass/DeepRecommender.py:24) 
 u_idx (defined at ../baseclass/DeepRecommender.py:19)

As you updated Deep Graph Collaborative Filtering recently, have you tried to implement another state-of-the-art algorithm GraphRec from the paper Graph Neural Network for Social Recommendation in RecQ?

I met some problems when implementing GraphRec as getting worse results than reported in the paper. I'm wondering if it's because my implementation is wrong or GraphRec doesn't work as reported.

于老师，您好！我最近在将ESRF的代码转为torch的版本，但是在计算可替代邻居的部分占用了太大的内存消耗，self.alternativeNeighborhood = tf.vectorized_map(fn=lambda em:getAlternativeNeighborhood(em),elems=user_features)，这句我转换为了循环函数，for elems in user_features: alternativeNeighborhood.append(getAlternativeNeighborhood(elems))，但是在用稍微大一点的数据集如Ciao时，都会内存溢出（如Ciao,7375个user，在为第700多个用户计算可替代邻居时内存就满了，8G的显存）。在打印GPU的使用情况发现，就是这句代码为每个用户计算时，占用了特别大内存消耗。然后我在这部分仅仅计算每个batch的users的可替代邻居，但是最后计算损失函数，又没有可替代邻居的embedding时，因此想请教老师，有什么能解决这句代码内存消耗的方法吗？

Hi. I have some problems with both the implementation and several formulas in your RSGAN paper.

1. As you present the whole min-max loss of RSGAN in Eq. 9, however, in Eq. 8, the generator's loss is defined without the latter part of Eq. 9, log(sigmoid(x_uz-x_uj)), which is the difference between the scores of generated items and negative items. I think, to reach an equilibrium, the loss of the generator should maximize the whole loss of Eq.9. Why do you design the generator loss without the latter part of Eq.9?
2. As in your implementation, the loss of the generator is defined without the log part and is multiplied by 30. Is it implemented in this way to avoid too small gradients brought by the log function? And is this so-called implementation trick by others?

Thanks for your time. Your reply always helps a lot.

我在lastfm上运行了代码中MHCN的模型，看到代码中实现的是s2-MHCN，但运行出的结果与论文中给出的实验数据有一定出入（如下： Precision:0.19607427055702917 Recall:0.1991364574123199 F1:0.197593500708107 NDCG:0.24458583342484325

Hello professor!thanks for your code. when I run main.py and choose the algorithm . it always has the same question: Traceback (most recent call last): File "C:/Users/Administrator/Desktop/新建文件夹/RecQ/RecQ-master/main/main.py", line 67, in recSys.execute() File "C:\Users\Administrator\Desktop\新建文件夹\RecQ\RecQ-master\main\RecQ.py", line 100, in execute self.measure = [dict.setdafault(m)[i] for i in range(1,k+1)] File "C:\Users\Administrator\Desktop\新建文件夹\RecQ\RecQ-master\main\RecQ.py", line 100, in self.measure = [dict.setdafault(m)[i] for i in range(1,k+1)] AttributeError: type object 'dict' has no attribute 'setdafault' How can I solve it?looking for your reply.

您好，我看到您写的LOCABAL算法了，我对您用的梯度下降公式有点疑问，您在哪儿找到的这个公式，我在原论文里没找到啊，然后我在其他论文里找到了部分，a synthetic approach for recommendation combining ratings social relations and reviews. 这个论文里使用了LOCABAL算法，他给出的迭代公式和您代码里写的还是有很大不同的。这里我还是有点疑惑的。

Hi, I've run LightGCN in QRec recently. But I find that the performance of it is not good. The results I get on Yelp are Precision:0.0252', 'Recall:0.0575', 'F1:0.0350', 'NDCG:0.0465 after 1000 epochs. While the results of SEPT on the same dataset are 'Precision:0.0310', 'Recall:0.0712', 'F1:0.0432', 'NDCG:0.0583'] after 30 epochs. And I also set if epoch > self.maxEpoch / 3: to if epoch > self.maxEpoch: in SEPT, which means SEPT will behave like LightGCN. But the results, which are ['Precision:0.0289', 'Recall:0.0641', 'F1:0.0399', 'NDCG:0.0532'] after 30 epochs, are also much better than the original one. BTW, the initial rec_loss of the degraded SEPT is much lower than the original LightGCN, ~300/batch vs ~1000/batch. So I want to know how to make the original LightGCN better. I've already tried the same normalization and initiliazation as SEPT, but it doesn't work.

Hi, I noticed that you are evaluating one user at each time. While my test data has hundreds of thousands users, and the evaluation will take about a few minutes. While the train on GPU is very fast. So can you provide some fast evaluation method? Thank you!

Initializing model [1]... Traceback (most recent call last): File "C:/Users/MACHENIKE/Desktop/论文代码/QRec-master/main.py", line 54, in recSys.execute() File "C:\Users\MACHENIKE\Desktop\论文代码\QRec-master\QRec.py", line 114, in execute eval(recommender).execute() File "C:\Users\MACHENIKE\Desktop\论文代码\QRec-master\base\recommender.py", line 214, in execute self.initModel() File "C:\Users\MACHENIKE\Desktop\论文代码\QRec-master\model\ranking\TBPR.py", line 41, in initModel self.t_s = self.weights[len(self.weights)/2+1:].sum()/(len(self.weights[len(self.weights)/2+1:])+0.0) TypeError: slice indices must be integers or None or have an index method