Dear, I found this github repository, because I want to extend on the topic of bundle generation and graph generation for recommender systems. To do this I wanted to test your bundle generation model (BGGN) on other datasets. Can I get access to the code for bundle generation?

在作者提供的dataset文件中给出了negative sample的采样步骤 `class BundleTrainDataset(BasicDataset):

def __init__(self, path, name, item_data, assist_data, seed=None):
    super().__init__(path, name, 'train', 1)
    # U-B
    self.U_B_pairs = self.load_U_B_interaction()
    indice = np.array(self.U_B_pairs, dtype=np.int32)
    values = np.ones(len(self.U_B_pairs), dtype=np.float32)
    
    self.ground_truth_u_b = sp.coo_matrix(
        (values, (indice[:, 0], indice[:, 1])), shape=(self.num_users, self.num_bundles)).tocsr()
    

    print_statistics(self.ground_truth_u_b, 'U-B statistics in train')

   
    if CONFIG['sample'] == 'hard': 
        #  1. u_p --> b_n1
        u_b_from_i = item_data.ground_truth_u_i @ assist_data.ground_truth_b_i.T 
        u_b_from_i = u_b_from_i.todense() 
        bn1_window = [int(i*self.num_bundles) for i in CONFIG['hard_window']] 
        #'hard_window': [0.7, 1.0]
        self.u_b_for_neg_sample = np.argsort(u_b_from_i, axis=1)[:,bn1_window[0]:bn1_window[1]]
        
        
        #  2. b_p --> b_n2
        overlap_graph = assist_data.ground_truth_b_i @ assist_data.ground_truth_b_i.T
        overlap_graph = overlap_graph.todense()
        bn2_window = [int(i*self.num_bundles) for i in CONFIG['hard_window']]
        self.b_b_for_neg_sample = np.argsort(overlap_graph, axis=1)[:,bn2_window[0]:bn2_window[1]]
        #对每个bundle i，选出与其重合度较高的一些bundle

def __getitem__(self, index):
    user_b, pos_bundle = self.U_B_pairs[index]
    all_bundles = [pos_bundle]
    if CONFIG['sample'] == 'simple': #普通模式
        while True:
            i = np.random.randint(self.num_bundles)
            if self.ground_truth_u_b[user_b, i] == 0 and not i in all_bundles:
                all_bundles.append(i)
                if len(all_bundles) == self.neg_sample+1:
                    break
    elif CONFIG['sample'] == 'hard': #hard模式
        hard_probability = round(np.random.uniform(0, 1), 1)
        if  hard_probability <= CONFIG['hard_prob'][0]:
            while True:
                i = np.random.randint(self.u_b_for_neg_sample.shape[1])
                b_n1 = self.u_b_for_neg_sample[user_b, i]
                if self.ground_truth_u_b[user_b, b_n1] == 0 and not b_n1 in all_bundles:
                    all_bundles.append(b_n1)
                    if len(all_bundles) == self.neg_sample+1:
                        break
        elif CONFIG['hard_prob'][0] < hard_probability \
            <= CONFIG['hard_prob'][0] + CONFIG['hard_prob'][1]:
            while True:
                i = np.random.randint(self.b_b_for_neg_sample.shape[1])
                b_n2 = self.b_b_for_neg_sample[pos_bundle, i]
                if self.ground_truth_u_b[user_b, b_n2] == 0 and not b_n2 in all_bundles:
                    all_bundles.append(b_n2)
                    if len(all_bundles) == self.neg_sample+1:
                        break
        else:
            while True:
                i = np.random.randint(self.num_bundles)
                if self.ground_truth_u_b[user_b, i] == 0 and not i in all_bundles:
                    all_bundles.append(i)
                    if len(all_bundles) == self.neg_sample+1:
                        break
    else:
        raise ValueError(r"sample's method is wrong")

    return torch.LongTensor([user_b]), torch.LongTensor(all_bundles)

` 代码里使用np.argsort和hard_window来取出交互程度为hard_window%的bundle，但实际上np.argsort返回的是每个项从小到大排列时的位置索引，并非返回排序后的列表；所以我认为在后续的__getitem__函数抽样取出的样本并非hard sample

• Futher Train

Replace 'sample' from 'simple' to 'hard' in CONFIG and add model file path obtained by Train to 'conti_train', then run

python main.py

I cannot find the model file path obtained by Train, could you tell me where i can find it?

Thank you very much!

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "E:\Anconda3\lib\multiprocessing\spawn.py", line 116, in spawn_main
Traceback (most recent call last):
  File "main.py", line 155, in <module>
    exitcode = _main(fd, parent_sentinel)
  File "E:\Anconda3\lib\multiprocessing\spawn.py", line 126, in _main
    self = reduction.pickle.load(from_parent)
_pickle.UnpicklingError: pickle data was truncated
    main()
  File "main.py", line 120, in main
    trainloss = train(model, epoch+1, train_loader, op, device, CONFIG, loss_func
)
  File "F:\GNN\BGCN-master\train.py", line 15, in train
    for i, data in enumerate(loader):
  File "E:\Anconda3\lib\site-packages\torch\utils\data\dataloader.py", line 352,
in __iter__
    return self._get_iterator()
  File "E:\Anconda3\lib\site-packages\torch\utils\data\dataloader.py", line 294,
in _get_iterator
    return _MultiProcessingDataLoaderIter(self)
  File "E:\Anconda3\lib\site-packages\torch\utils\data\dataloader.py", line 801,
in __init__
    w.start()
  File "E:\Anconda3\lib\multiprocessing\process.py", line 121, in start
    self._popen = self._Popen(self)
  File "E:\Anconda3\lib\multiprocessing\context.py", line 224, in _Popen
    return _default_context.get_context().Process._Popen(process_obj)
  File "E:\Anconda3\lib\multiprocessing\context.py", line 327, in _Popen
    return Popen(process_obj)
  File "E:\Anconda3\lib\multiprocessing\popen_spawn_win32.py", line 93, in __init__
    reduction.dump(process_obj, to_child)
  File "E:\Anconda3\lib\multiprocessing\reduction.py", line 60, in dump
    ForkingPickler(file, protocol).dump(obj)
OSError: [Errno 22] Invalid argument

您好，这是我在train sample后接着训练hard报出的错，请问一下是什么问题呢？万分感谢