Fix bugs in transform

data/transform.py

@@ -5,22 +5,22 @@
 
 
 def construct_transform(config):
-    if config['transform'] is None:
+    if config['unisrec_transform'] is None:
         logger = getLogger()
         logger.warning('Equal transform')
         return Equal(config)
     else:
         str2transform = {
             'plm_emb': PLMEmb
         }
-        return str2transform[config['transform']](config)
+        return str2transform[config['unisrec_transform']](config)
 
 
 class Equal:
     def __init__(self, config):
         pass
 
-    def __call__(self, dataloader, interaction):
+    def __call__(self, dataset, interaction):
         return interaction
 
 
@@ -31,13 +31,13 @@ def __init__(self, config):
         self.item_drop_ratio = config['item_drop_ratio']
         self.item_drop_coefficient = config['item_drop_coefficient']
 
-    def __call__(self, dataloader, interaction):
+    def __call__(self, dataset, interaction):
         '''Sequence augmentation and PLM embedding fetching
         '''
         item_seq_len = interaction['item_length']
         item_seq = interaction['item_id_list']
 
-        plm_embedding = dataloader.dataset.plm_embedding
+        plm_embedding = dataset.plm_embedding
         item_emb_seq = plm_embedding(item_seq)
         pos_item_id = interaction['item_id']
         pos_item_emb = plm_embedding(pos_item_id)
@@ -59,7 +59,7 @@ def __call__(self, dataloader, interaction):
             item_emb_seq_aug = plm_embedding(item_seq_aug)
         else:
             # Word drop
-            plm_embedding_aug = dataloader.dataset.plm_embedding_aug
+            plm_embedding_aug = dataset.plm_embedding_aug
             full_item_emb_seq_aug = plm_embedding_aug(item_seq)
 
             item_seq_aug = item_seq

props/finetune.yaml

@@ -4,7 +4,7 @@ alias_of_item_id: [item_id_list]
 load_col:
   inter: [user_id, item_id_list, item_id]
 train_neg_sample_args: ~
-transform: ~
+unisrec_transform: ~
 
 topk: [10, 50]
 metrics: [HIT, NDCG]

props/pretrain.yaml

@@ -3,8 +3,8 @@ benchmark_filename: [train]
 alias_of_item_id: [item_id_list]
 load_col:
   inter: [user_id, item_id_list, item_id]
-neg_sampling: ~
-transform: plm_emb
+train_neg_sample_args: ~
+unisrec_transform: plm_emb
 
 train_stage: pretrain
 pretrain_epochs: 300