FIX: add base.yaml and fix tmtf

debias_config/base.yaml

@@ -0,0 +1,58 @@
+
+# general
+gpu_id: 0
+use_gpu: True
+seed: 2020
+state: INFO
+reproducibility: True
+data_path: 'dataset/'
+checkpoint_dir: 'saved'
+show_progress: True
+save_dataset: False
+dataset_save_path: ~
+save_dataloaders: False
+dataloaders_save_path: ~
+log_wandb: False
+wandb_project: 'recbole'
+LABEL_FIELD : 'rating'
+RATING_FIELD : 'rating'
+TIME_FIELD: 'wday'
+#WDAY_FIELD: 'wday'
+# training settings
+epochs: 100
+train_batch_size: 1024
+learner: adam
+learning_rate: 0.001
+neg_sampling:
+  uniform: 10
+
+eval_step: 1
+stopping_step: 50
+clip_grad_norm: ~
+# clip_grad_norm:  {'max_norm': 5, 'norm_type': 2}
+weight_decay: 0.0
+loss_decimal_place: 4
+require_pow: False
+
+# evaluation settings
+eval_args: 
+  split: {'RS':[0.8,0.1,0.1]}
+  group_by: user
+  order: TO
+  mode: labeled
+repeatable: False
+metrics: ["MSE","RMSE","MAE","AUC","LogLoss"]
+topk: [10]
+load_col:  {'inter': ['user_id', 'item_id','rating','timestamp','wday']}
+valid_metric: rmse
+valid_metric_bigger: True
+eval_batch_size: 2048
+metric_decimal_place: 4
+normalize_field: ['timestamp','rating']
+normalize_all: ~
+K: 7
+T: 100
+M: 3
+sig: 0.5
+embedding_size: 64
+benchmark_filename: ['train','valid','test']

debias_config/get_dps.yaml

@@ -63,4 +63,4 @@ gamma_v: 1.1
 gamma_t: 1.1
 task: 'ps' #ps,ips,dr,dips
 psv_path: 'init_ps/item_ps_week.pth'
-pst_path: 'init_ps/tmtf_time_ps_week.pth'
+pst_path: 'init_ps/tmtf_time_ps_global.pth'

debias_config/get_ps.yaml

@@ -62,5 +62,5 @@ base_freq: 1
 gamma_v: 1.1
 gamma_t: 1.1
 task: 'ps' #ps,ips,dr,dips
-psv_path: 'init_ps/item_ps_week.pth'
-pst_path: 'init_ps/time_ps_week.pth'
+psv_path: 'init_ps/item_ps_global.pth'
+pst_path: 'init_ps/time_ps_global.pth'

recbole/model/general_recommender/tmtf.py

@@ -32,7 +32,6 @@ def __init__(self, config, dataset):
         self.LABEL = config['LABEL_FIELD']
         self.RATING = config['RATING_FIELD']
         self.TIME = config['TIME_FIELD']
-        self.WDAY = config['WDAY_FIELD']
         # load parameters info
         self.embedding_size = config['embedding_size']
         self.K = config['K']
@@ -73,7 +72,7 @@ def calculate_loss(self, interaction, weight=None):
         user = interaction[self.USER_ID]
         item = interaction[self.ITEM_ID]
         label = interaction[self.LABEL]
-        time = interaction[self.WDAY].long()
+        time = interaction[self.TIME].long()
         pred = self.forward(user, item, time)
 
         loss = self.loss_fct(pred, label)
@@ -83,7 +82,7 @@ def calculate_loss(self, interaction, weight=None):
     def predict(self, interaction):
         user = interaction[self.USER_ID]
         item = interaction[self.ITEM_ID]
-        time = interaction[self.WDAY]
+        time = interaction[self.TIME].long()
         pred = self.forward(user, item, time)
         return pred