Merge pull request #2 from RUCAIBox/master

up to date

main.py

@@ -1,16 +1,15 @@
 from logging import getLogger
 from recbox.config import Config
-from recbox.data import Dataset, data_preparation
-from recbox.model.general_recommender.bprmf import BPRMF
-from recbox.trainer import Trainer
+from recbox.data import create_dataset, data_preparation
+from recbox.trainer import get_trainer
 from recbox.utils import init_logger, get_model
 
 config = Config('properties/overall.config')
 config.init()
 init_logger(config)
 logger = getLogger()
 
-dataset = Dataset(config)
+dataset = create_dataset(config)
 logger.info(dataset)
 
 # If you want to customize the evaluation setting,
@@ -20,7 +19,7 @@
 model = get_model(config['model'])(config, train_data).to(config['device'])
 logger.info(model)
 
-trainer = Trainer(config, model)
+trainer = get_trainer(config['MODEL_TYPE'])(config, model)
 
 # trainer.resume_checkpoint('saved/model_best.pth')
 best_valid_score, _ = trainer.fit(train_data, valid_data)

properties/dataset/amazon-game.config

@@ -3,6 +3,7 @@
 ########define the UIRT columns
 USER_ID_FIELD='reviewerID'
 ITEM_ID_FIELD='asin'
+RATING_FIELD='overall'
 NEG_PREFIX='neg_'
 LABEL_FIELD='label'
 

properties/dataset/kgdata_example.config

@@ -0,0 +1,22 @@
+[data]
+
+########define the UIRT columns
+USER_ID_FIELD='user_id'
+ITEM_ID_FIELD='item_id'
+NEG_PREFIX='neg_'
+HEAD_ENTITY_ID_FIELD='head_id'
+TAIL_ENTITY_ID_FIELD='tail_id'
+RELATION_ID_FIELD='relation_id'
+ENTITY_ID_FIELD='entity_id'
+LABEL_FIELD='label'
+
+#########select load columns
+load_col={'inter': ['user_id', 'item_id', 'rating'], 'kg': ['head_id', 'relation_id', 'tail_id'], 'link': ['item_id', 'entity_id']}
+
+########data separator
+field_separator='\t'
+seq_separator=' '
+
+########data filter
+lowest_val={'rating':3}
+drop_filter_field=True

properties/dataset/ml-100k.config

@@ -5,6 +5,7 @@ USER_ID_FIELD='user_id'
 ITEM_ID_FIELD='item_id'
 NEG_PREFIX='neg_'
 LABEL_FIELD='label'
+RATING_FIELD='rating'
 
 #########select load columns
 # USER_ID_FIELD & ITEM_ID_FIELD can be omitted

properties/dataset/ml-1m.config

@@ -5,6 +5,7 @@ USER_ID_FIELD='user_id'
 ITEM_ID_FIELD='item_id'
 NEG_PREFIX='neg_'
 LABEL_FIELD='label'
+RATING_FIELD='rating'
 
 #########select load columns
 load_col={'inter': ['user_id', 'item_id', 'rating']}

properties/dataset/yelp.config

@@ -5,6 +5,7 @@ USER_ID_FIELD='user_id'
 ITEM_ID_FIELD='business'
 NEG_PREFIX='neg_'
 LABEL_FIELD='label'
+RATING_FIELD='stars'
 
 #########select load columns
 load_col={'inter': ['user_id', 'business', 'stars']}

properties/model/DMF.config

@@ -1,5 +1,8 @@
 [model]
 
-layers = [64,64]
-reg_weight = 0.0
+# The dimensions of the last layer of users and items must be the same
+inter_matrix_type = '01'
+# inter_matrix_type = 'rating'  
+user_layers_dim = [64, 64]
+item_layers_dim = [64, 64]
 min_y_hat = 1e-6

properties/model/FPMC.config

@@ -0,0 +1,11 @@
+[model]
+
+MAX_ITEM_LIST_LENGTH=50
+TARGET_PREFIX='target_'
+LIST_SUFFIX='_list'
+ITEM_LIST_LENGTH_FIELD='item_length'
+USER_ID_FIELD='user_id'
+NEG_PREFIX='neg_'
+
+embedding_size=64
+neg_count=1

properties/model/GRU4Rec.config

@@ -3,9 +3,9 @@
 MAX_ITEM_LIST_LENGTH=50
 TARGET_PREFIX='target_'
 LIST_SUFFIX='_list'
-POSITION_FIELD='position_id'
 ITEM_LIST_LENGTH_FIELD='item_length'
 
 embedding_size=64
+hidden_size=32
 num_layers=1
 dropout=0

properties/model/LightGCN.config

@@ -0,0 +1,5 @@
+[model]
+
+embedding_size=64
+layers=3
+delay=1e-4

properties/model/NARM.config

@@ -0,0 +1,11 @@
+[model]
+
+MAX_ITEM_LIST_LENGTH=50
+TARGET_PREFIX='target_'
+LIST_SUFFIX='_list'
+ITEM_LIST_LENGTH_FIELD='item_length'
+
+embedding_size=64
+hidden_size=32
+n_layers=1
+dropout=[0.25,0.5]

properties/overall.config

@@ -17,7 +17,9 @@ eval_setting=RO_RS, full
 group_by_user=True
 split_ratio=[0.8, 0.1, 0.1]
 # leave_one_num=2
-real_time_neg_sampling=True
+
+# DataLoader args
+real_time_process=True
 
 # training
 epochs=400
@@ -29,6 +31,7 @@ eval_step=1
 valid_metric=MRR@10
 stopping_step=10
 seed=2020
+training_neg_sample_num=1
 
 # evaluating
 metrics=["Recall", "MRR","NDCG"]

recbox/config/eval_setting.py

@@ -3,9 +3,9 @@
 # @Email  : [email protected]
 
 # UPDATE:
-# @Time   : 2020/8/19 18:56
-# @Author : Yupeng Hou
-# @Email  : [email protected]
+# @Time   : 2020/8/19 18:56, 2020/8/31
+# @Author : Yupeng Hou, Yushuo Chen
+# @Email  : [email protected], [email protected]
 
 
 class EvalSetting(object):
@@ -23,12 +23,29 @@ def __init__(self, config):
                 setattr(self, args, config[args])
 
     def __str__(self):
-        info = 'Evaluation Setting:\n'
-        info += ('\tGroup by {}\n'.format(self.group_field) if self.group_field is not None else '\tNo Grouping\n')
-        info += ('\tOrdering: {}\n'.format(self.ordering_args) if (self.ordering_args is not None and self.ordering_args['strategy'] != 'none') else '\tNo Ordering\n')
-        info += ('\tSplitting: {}\n'.format(self.split_args) if (self.split_args is not None and self.split_args['strategy'] != 'none') else '\tNo Splitting\n')
-        info += ('\tNegative Sampling: {}'.format(self.neg_sample_args) if (self.neg_sample_args is not None and self.neg_sample_args['strategy'] != 'none') else '\tNo Negative Sampling\n')
-        return info
+        info = ['Evaluation Setting:']
+
+        if self.group_field:
+            info.append('Group by {}'.format(self.group_field))
+        else:
+            info.append('No Grouping')
+
+        if self.ordering_args is not None and self.ordering_args['strategy'] != 'none':
+            info.append('Ordering: {}'.format(self.ordering_args))
+        else:
+            info.append('No Ordering')
+
+        if self.split_args is not None and self.split_args['strategy'] != 'none':
+            info.append('Splitting: {}'.format(self.split_args))
+        else:
+            info.append('No Splitting')
+
+        if self.neg_sample_args is not None and self.neg_sample_args['strategy'] != 'none':
+            info.append('Negative Sampling: {}'.format(self.neg_sample_args))
+        else:
+            info.append('No Negative Sampling')
+
+        return '\n\t'.join(info)
 
     def __repr__(self):
         return self.__str__()
@@ -135,24 +152,23 @@ def split_by_value(self, field, values, ascending=True):
     Args:
         strategy (str): Either 'none', 'full' or 'by'.
         by (int): Negative Sampling `by` neg cases for one pos case.
-        real_time (bool): real time negative sampling if True, else negative cases will be pre-sampled and stored.
         distribution (str): distribution of sampler, either 'uniform' or 'popularity'.
 
     Example:
-        >>> es.neg_sample_to(100, real_time=True)
+        >>> es.neg_sample_to(100)
         >>> es.neg_sample_by(1)
     """
-    def set_neg_sampling(self, strategy='none', real_time=False, distribution='uniform', **kwargs):
+    def set_neg_sampling(self, strategy='none', distribution='uniform', **kwargs):
         legal_strategy = {'none', 'full', 'by'}
         if strategy not in legal_strategy:
             raise ValueError('Negative Sampling Strategy [{}] should in {}'.format(strategy, list(legal_strategy)))
         if strategy == 'full' and distribution != 'uniform':
             raise ValueError('Full Sort can not be sampled by distribution [{}]'.format(distribution))
-        self.neg_sample_args = {'strategy': strategy, 'real_time': real_time, 'distribution': distribution}
+        self.neg_sample_args = {'strategy': strategy, 'distribution': distribution}
         self.neg_sample_args.update(kwargs)
 
-    def neg_sample_by(self, by, real_time=False, distribution='uniform'):
-        self.set_neg_sampling(strategy='by', by=by, real_time=real_time, distribution=distribution)
+    def neg_sample_by(self, by, distribution='uniform'):
+        self.set_neg_sampling(strategy='by', by=by, distribution=distribution)
 
     r"""Presets
 
@@ -161,13 +177,13 @@ def neg_sample_by(self, by, real_time=False, distribution='uniform'):
     full: all non-ground-truth items
     uni: uniform sampling       pop: popularity sampling        neg_sample_by 100 by default.
     """
-    def RO_RS(self, ratios=[0.8, 0.1, 0.1], group_by_user=True):
+    def RO_RS(self, ratios=(0.8, 0.1, 0.1), group_by_user=True):
         if group_by_user:
             self.group_by_user()
         self.random_ordering()
         self.split_by_ratio(ratios)
 
-    def TO_RS(self, ratios=[0.8, 0.1, 0.1], group_by_user=True):
+    def TO_RS(self, ratios=(0.8, 0.1, 0.1), group_by_user=True):
         if group_by_user:
             self.group_by_user()
         self.temporal_ordering()
@@ -185,17 +201,17 @@ def TO_LS(self, leave_one_num=1, group_by_user=True):
         self.temporal_ordering()
         self.leave_one_out(leave_one_num=leave_one_num)
 
-    def uni100(self, real_time=False):
-        self.neg_sample_by(100, real_time=real_time)
+    def uni100(self):
+        self.neg_sample_by(100)
 
-    def pop100(self, real_time=False):
-        self.neg_sample_by(100, real_time=real_time, distribution='popularity')
+    def pop100(self):
+        self.neg_sample_by(100, distribution='popularity')
 
-    def uni1000(self, real_time=False):
-        self.neg_sample_by(1000, real_time=real_time)
+    def uni1000(self):
+        self.neg_sample_by(1000)
 
-    def pop1000(self, real_time=False):
-        self.neg_sample_by(1000, real_time=real_time, distribution='popularity')
+    def pop1000(self):
+        self.neg_sample_by(1000, distribution='popularity')
 
-    def full(self, real_time=True):
-        self.set_neg_sampling(strategy='full', real_time=real_time)
+    def full(self):
+        self.set_neg_sampling(strategy='full')