update files

experiment/run_ddpg.py

@@ -11,14 +11,10 @@
 import torch.backends.cudnn as cudnn
 
 import sys
-# sys.path.append('/home/xinghua/Hongyang/Code-submit')
-sys.path.append('/home/workshop/lhy/code-submit')
 
 from collections import defaultdict
-# from memory_profiler import profile
 from tqdm import tqdm
 from daisy.model.KNNCFRecommender import ItemKNNCF
-# from daisy.model.TRDRecommender import TRD, get_next_state, get_reward, create_initial_state
 from daisy.utils.loader import load_rate, split_test, get_ur, PairMFData, get_ur_l
 from daisy.utils.metrics import precision_at_k, recall_at_k, map_at_k, hr_at_k, mrr_at_k, ndcg_at_k
 
@@ -35,13 +31,13 @@ def __init__(self, user_num, item_num, factor_num, pretrain_model, pretrain):
 
         if pretrain:
             if method == 'Item2Vec':
-                item_weights = np.random.normal(size=(item_num + 1, factor_num), scale=0.01)
+                item_weights = np.zeros((item_num + 1, factor_num))
                 for k, v in model.item2vec.items():
                     if isinstance(k, int):
                         item_weights[k] = v
                 self.embed_item.weight.data.copy_(torch.from_numpy(item_weights))
 
-                user_weights = np.random.normal(size=(user_num, factor_num), scale=0.01)
+                user_weights = np.zeros((user_num, factor_num))
                 for k, v in model.user_vec_dict.items():
                     if isinstance(k, int):
                         user_weights[k] = v
@@ -51,18 +47,14 @@ def __init__(self, user_num, item_num, factor_num, pretrain_model, pretrain):
 
             if method == 'bprmf':
                 weight = model.embed_item.weight.cpu().detach()
-                pad = np.random.normal(size=(1, factor_num), scale=0.01)
+                pad = np.zeros((1, factor_num))
                 pad = torch.FloatTensor(pad)
                 weight = torch.cat([weight, pad])
                 self.embed_item.weight.data.copy_(weight) 
 
                 weight = model.embed_user.weight.cpu().detach()
                 self.embed_user.weight.data.copy_(weight)
-                print("embedding load completed")
-        else:
-            print("not load item2vec")
-            nn.init.normal_(self.embed_user.weight, std=0.01)
-            nn.init.normal_(self.embed_item.weight, std=0.01)              
+                print("embedding load completed")           
 
 class Actor(nn.Module):
     def __init__(self, embedding, factor_num=32, gpuid='0'):

experiment/run_dqn.py

@@ -11,14 +11,10 @@
 import torch.backends.cudnn as cudnn
 
 import sys
-sys.path.append('/home/xinghua/Hongyang/Code-submit')
-# sys.path.append('/home/workshop/lhy/code-submit')
 
 from collections import defaultdict
-# from memory_profiler import profile
 from tqdm import tqdm
 from daisy.model.KNNCFRecommender import ItemKNNCF
-# from daisy.model.TRDRecommender import TRD, get_next_state, get_reward, create_initial_state
 from daisy.utils.loader import load_rate, split_test, get_ur, PairMFData, get_ur_l
 from daisy.utils.metrics import precision_at_k, recall_at_k, map_at_k, hr_at_k, mrr_at_k, ndcg_at_k
 
@@ -38,31 +34,29 @@ def __init__(self, user_num, item_num, model, method, factor_num=32, pretrain=1,
 
         if self.pretrain:
             if method == 'Item2Vec':
-                item_weights = np.random.normal(size=(item_num + 1, factor_num), scale=0.01)
+                item_weights = np.zeros((item_num + 1, factor_num))
                 for k, v in model.item2vec.items():
                     if isinstance(k, int):
                         item_weights[k] = v
                 self.embed_item.weight.data.copy_(torch.from_numpy(item_weights))
 
-                user_weights = np.random.normal(size=(user_num, factor_num), scale=0.01)
+                user_weights = np.zeros((user_num, factor_num))
                 for k, v in model.user_vec_dict.items():
                     if isinstance(k, int):
                         user_weights[k] = v
                 self.embed_user.weight.data.copy_(torch.from_numpy(user_weights))
                 del item_weights, user_weights
             if method == 'bprmf':
                 weight = model.embed_item.weight.cpu().detach()
-                pad = np.random.normal(size=(1, factor_num), scale=0.01)
+                pad = np.zeros((1, factor_num))
                 pad = torch.FloatTensor(pad)
                 weight = torch.cat([weight, pad])
                 self.embed_item.weight.data.copy_(weight) 
 
                 weight = model.embed_user.weight.cpu().detach()
                 self.embed_user.weight.data.copy_(weight)
                 print("embedding load completed")
-        else:
-            nn.init.normal_(self.embed_user.weight, std=0.01)
-            nn.init.normal_(self.embed_item.weight, std=0.01)
+
 
         self.fc1 = nn.Linear(self.factor_num * 7, 256)
         self.fc1.weight.data.normal_(0,0.1)

experiment/run_item2Vec.py

@@ -11,8 +11,6 @@
 import torch.utils.data as data
 
 import sys
-sys.path.append('/home/xinghua/Hongyang/Code-submit')
-# sys.path.append('/home/workshop/lhy/code-submit')
 
 from daisy.model.Item2VecRecommender import Item2Vec
 from daisy.utils.loader import load_rate, split_test, get_ur, BuildCorpus, PermutedSubsampledCorpus

experiment/run_item2Vec_train.py

@@ -11,8 +11,6 @@
 import torch.utils.data as data
 
 import sys
-sys.path.append('/home/xinghua/Hongyang/Code-submit')
-# sys.path.append('/home/workshop/lhy/code-submit')
 
 from daisy.model.Item2VecRecommender import Item2Vec
 from daisy.utils.loader import load_rate, split_test, get_ur, BuildCorpus, PermutedSubsampledCorpus

experiment/run_itemknncf.py

@@ -10,8 +10,7 @@
 import torch.utils.data as data
 
 import sys
-# sys.path.append('/home/xinghua/Hongyang/Code-submit')
-sys.path.append('/home/workshop/lhy/code-submit')
+
 
 from daisy.model.KNNCFRecommender import ItemKNNCF
 from daisy.utils.loader import load_rate, split_test, get_ur
@@ -51,9 +50,6 @@
                         help='The (min) number of neighbors to take into account')
     args = parser.parse_args()
 
-    '''Test Process for Metrics Exporting'''
-    # df, user_num, item_num = load_rate(args.dataset, args.prepro, binary=False)
-    # train_set, test_set = split_test(df, args.test_method, args.test_size)
 
     #temporary used for tuning test result
     train_set1 = pd.read_csv(f'../experiment_data/train1_{args.dataset}_{args.prepro}.dat')
@@ -83,7 +79,6 @@
 
     # get ground truth
     test_ur = get_ur(test_set)
-    # validate_ur = get_ur(validate_set)
     total_train_ur = get_ur(train_set)
 
     # initial candidate item pool

experiment/run_itemknncf_train.py

@@ -10,8 +10,6 @@
 import torch.utils.data as data
 
 import sys
-sys.path.append('/home/xinghua/Hongyang/Code-submit')
-# sys.path.append('/home/workshop/lhy/code-submit')
 
 from daisy.model.KNNCFRecommender import ItemKNNCF
 from daisy.utils.loader import load_rate, split_test, get_ur
@@ -51,9 +49,6 @@
                         help='The (min) number of neighbors to take into account')
     args = parser.parse_args()
 
-    '''Test Process for Metrics Exporting'''
-    # df, user_num, item_num = load_rate(args.dataset, args.prepro, binary=False)
-    # train_set, test_set = split_test(df, args.test_method, args.test_size)
 
     #temporary used for tuning test result
     train_set1 = pd.read_csv(f'../experiment_data/train1_{args.dataset}_{args.prepro}.dat')

experiment/run_mostpop.py

@@ -7,19 +7,12 @@
 from tqdm import tqdm
 
 import sys
-# sys.path.append('/home/xinghua/Hongyang/Code-submit')
-sys.path.append('/home/workshop/lhy/code-submit')
 
 from daisy.model.MostPopRecommender import MostPop
 from daisy.utils.loader import load_rate, split_test, get_ur
 from daisy.utils.metrics import precision_at_k, recall_at_k, map_at_k, hr_at_k, mrr_at_k, ndcg_at_k
 
 if __name__ == '__main__':
-    # load logger configuration
-    # logging.config.fileConfig("./log.conf")
-    # logger_name = 'MostPop'
-    # logger = logging.getLogger(logger_name)
-    # logger.debug('MostPop experiment running...')
 
     parser = argparse.ArgumentParser(description='Most-Popular recommender test')
     # common settings
@@ -45,10 +38,6 @@
                         help='Initial selected number of Most-popular')
     args = parser.parse_args()
 
-    '''Test Process for Metrics Exporting'''
-    # df, user_num, item_num = load_rate(args.dataset, args.prepro)
-    # train_set, test_set = split_test(df, args.test_method, args.test_size)
-
     #temporary used for tuning test result
     train_set1 = pd.read_csv(f'../experiment_data/train1_{args.dataset}_{args.prepro}.dat')
     train_set2 = pd.read_csv(f'../experiment_data/train2_{args.dataset}_{args.prepro}.dat')
@@ -57,10 +46,8 @@
 
     train_set1['rating'] = 1.0
     train_set2['rating'] = 1.0
-    # validate_set['rating'] = 1.0
     test_set['rating'] = 1.0
     train_set = pd.concat([train_set1, train_set2], ignore_index=True)
-    # df = pd.concat([train_set, validate_set, test_set], ignore_index=True)
 
     split_idx = len(train_set)
     df = pd.concat([train_set,  test_set], ignore_index=True)

experiment/run_mostpop_train.py

@@ -8,19 +8,14 @@
 from tqdm import tqdm
 
 import sys
-sys.path.append('/home/xinghua/Hongyang/Code-submit')
-# sys.path.append('/home/workshop/lhy/code-submit')
+
 
 from daisy.model.MostPopRecommender import MostPop
 from daisy.utils.loader import load_rate, split_test, get_ur
 from daisy.utils.metrics import precision_at_k, recall_at_k, map_at_k, hr_at_k, mrr_at_k, ndcg_at_k
 
 if __name__ == '__main__':
-    # load logger configuration
-    # logging.config.fileConfig("./log.conf")
-    # logger_name = 'MostPop'
-    # logger = logging.getLogger(logger_name)
-    # logger.debug('MostPop experiment running...')
+
 
     parser = argparse.ArgumentParser(description='Most-Popular recommender test')
     # common settings
@@ -46,11 +41,7 @@
                         help='Initial selected number of Most-popular')
     args = parser.parse_args()
 
-    '''Test Process for Metrics Exporting'''
-    # df, user_num, item_num = load_rate(args.dataset, args.prepro)
-    # train_set, test_set = split_test(df, args.test_method, args.test_size)
 
-    #temporary used for tuning test result
     train_set1 = pd.read_csv(f'../experiment_data/train1_{args.dataset}_{args.prepro}.dat')
     train_set2 = pd.read_csv(f'../experiment_data/train2_{args.dataset}_{args.prepro}.dat')
 

experiment/run_pair_mf.py

@@ -10,8 +10,7 @@
 import torch.utils.data as data
 
 import sys
-# sys.path.append('/home/xinghua/Hongyang/Code-submit')
-sys.path.append('/home/workshop/lhy/code-submit')
+
 from daisy.model.pairwise.MFRecommender import PairMF
 from daisy.utils.loader import load_rate, split_test, get_ur, PairMFData
 from daisy.utils.metrics import precision_at_k, recall_at_k, map_at_k, hr_at_k, mrr_at_k, ndcg_at_k
@@ -81,18 +80,13 @@
                         help='save model or not')
     args = parser.parse_args()
 
-    '''Test Process for Metrics Exporting'''
-    # df, user_num, item_num = load_rate(args.dataset, args.prepro)
-    # train_set, test_set = split_test(df, args.test_method, args.test_size)
 
-    #temporary used for tuning test result
+
     train_set1 = pd.read_csv(f'../experiment_data/train1_{args.dataset}_{args.prepro}.dat')
     train_set2 = pd.read_csv(f'../experiment_data/train2_{args.dataset}_{args.prepro}.dat')
     # validate_set = pd.read_csv(f'../experiment_data/validate_{args.dataset}_{args.prepro}_{args.test_method}.dat')
     test_set = pd.read_csv(f'../experiment_data/test_{args.dataset}_{args.prepro}.dat')
-    # if args.dataset in ['yelp']:
-    #     train_set['timestamp'] = pd.to_datetime(train_set['timestamp'])
-    #     test_set['timestamp'] = pd.to_datetime(test_set['timestamp'])
+
     train_set1['rating'] = 1.0
     train_set2['rating'] = 1.0
     # validate_set['rating'] = 1.0

experiment/run_pair_mf_train.py

@@ -10,8 +10,6 @@
 import torch.utils.data as data
 
 import sys
-# sys.path.append('/home/xinghua/Hongyang/Code-submit')
-sys.path.append('/home/workshop/lhy/code-submit')
 
 from daisy.model.pairwise.MFRecommender import PairMF
 from daisy.utils.loader import load_rate, split_test, get_ur, PairMFData

experiment/run_pair_neumf.py

@@ -10,8 +10,6 @@
 import torch.utils.data as data
 
 import sys
-# sys.path.append('/home/xinghua/Hongyang/Code-submit')
-sys.path.append('/home/workshop/lhy/code-submit')
 
 from daisy.model.pairwise.NeuMFRecommender import PairNeuMF
 from daisy.utils.loader import load_rate, split_test, get_ur, PairMFData
@@ -91,11 +89,7 @@
                         help='gpu card ID')
     args = parser.parse_args()
 
-    '''Test Process for Metrics Exporting'''
-    # df, user_num, item_num = load_rate(args.dataset, args.prepro)
-    # train_set, test_set = split_test(df, args.test_method, args.test_size)
 
-    # temporary used for tuning test result
     train_set1 = pd.read_csv(f'../experiment_data/train1_{args.dataset}_{args.prepro}.dat')
     train_set2 = pd.read_csv(f'../experiment_data/train2_{args.dataset}_{args.prepro}.dat')
 

experiment/run_pair_neumf_train.py

@@ -10,8 +10,6 @@
 import torch.utils.data as data
 
 import sys
-sys.path.append('/home/xinghua/Hongyang/Code-submit')
-# sys.path.append('/home/workshop/lhy/code-submit')
 
 from daisy.model.pairwise.NeuMFRecommender import PairNeuMF
 from daisy.utils.loader import load_rate, split_test, get_ur, PairMFData
@@ -164,7 +162,6 @@
     test_ucands = defaultdict(list)
     for k, v in train2_ur.items():
         sample_num = candidates_num - len(v) if len(v) < candidates_num else 0
-        # sub_item_pool = item_pool - v - total_train_ur[k] - validate_ur[k]# remove GT & interacted
         sub_item_pool = item_pool - v - train1_ur[k] - test_ur[k]
         sample_num = min(len(sub_item_pool), sample_num)
         if sample_num == 0:

experiment/run_trd.py

@@ -11,8 +11,6 @@
 import torch.backends.cudnn as cudnn
 
 import sys
-# sys.path.append('/home/xinghua/Hongyang/Code-submit')
-sys.path.append('/home/workshop/lhy/code-submit')
 
 from collections import defaultdict
 from tqdm import tqdm
@@ -34,13 +32,13 @@ def __init__(self, user_num, item_num, model, method, factor_num=32, gpuid='0'):
         self.embed_item = nn.Embedding(item_num + 1, factor_num, padding_idx=item_num)
 
         if method == 'Item2Vec' or method == 'mostpop' or method == 'itemknn' :
-            item_weights = np.random.normal(size=(item_num + 1, factor_num), scale=0.01)
+            item_weights = np.zeros((item_num + 1, factor_num))
             for k, v in model.item2vec.items():
                 if isinstance(k, int):
                     item_weights[k] = v
             self.embed_item.weight.data.copy_(torch.from_numpy(item_weights))
 
-            user_weights = np.random.normal(size=(user_num, factor_num), scale=0.01)
+            user_weights = np.zeros((user_num, factor_num))
             for k, v in model.user_vec_dict.items():
                 if isinstance(k, int):
                     user_weights[k] = v
@@ -49,7 +47,7 @@ def __init__(self, user_num, item_num, model, method, factor_num=32, gpuid='0'):
             del item_weights, user_weights
         elif method == 'bprmf':
             weight = model.embed_item.weight.cpu().detach()
-            pad = np.random.normal(size=(1, factor_num), scale=0.01)
+            pad = np.zeros((1, factor_num))
             pad = torch.FloatTensor(pad)
             weight = torch.cat([weight, pad])
             self.embed_item.weight.data.copy_(weight) 
@@ -58,16 +56,14 @@ def __init__(self, user_num, item_num, model, method, factor_num=32, gpuid='0'):
             self.embed_user.weight.data.copy_(weight)
         elif method == 'neumf':
             weight = model.embed_item_GMF.weight.cpu().detach()
-            pad = np.random.normal(size=(1, factor_num), scale=0.01)
+            pad = np.zeros((1, factor_num))
             pad = torch.FloatTensor(pad)
             weight = torch.cat([weight, pad])
             self.embed_item.weight.data.copy_(weight) 
 
             weight = model.embed_user_GMF.weight.cpu().detach()
             self.embed_user.weight.data.copy_(weight)            
-        else: 
-            nn.init.normal_(self.embed_user.weight, std=0.01)
-            nn.init.normal_(self.embed_item.weight, std=0.01)
+
 
         if self.factor_num == 32:
             self.fc1 = nn.Linear(self.factor_num * 7, 256)