debiasing with different backbones (v1)

Author: pepsi2222

recstudio/data/__init__.py

@@ -1,4 +1,4 @@
-from recstudio.data.dataset import TripletDataset, SeqDataset, UserDataset, FullSeqDataset
+from recstudio.data.dataset import TripletDataset, SeqDataset, UserDataset, FullSeqDataset, DICEDataset, UBPRDataset
 from recstudio.data.advance_dataset import ALSDataset
 
 import os

recstudio/data/config/all.yaml

@@ -49,4 +49,9 @@ mapped_feat_field: [*u, *i]
 network_feat_field: [[[source_id:token, target_id:token]], [[head_id:token, tail_id:token, relation_id:token], [*i, entity_id:token]]]
 network_feat_header: [0, 0]
 
+# interactions that are missing completely at random
+mcar_feat_name: ~
+inter_feat_field: [*u, *i, *r, *t]
+inter_feat_header: ~
+
 save_cache: False # whether to save processed dataset to cache.

recstudio/data/config/amazon-beauty.yaml

@@ -22,7 +22,7 @@ field_separator: ","
 min_user_inter: 5
 min_item_inter: 5
 field_max_len: ~
-low_rating_threshold: 3.0
+low_rating_thres: 3.0
 max_seq_len: 50
 
 # network feature, including social network and knowledge graph, the first two fields are remapped the corresponding features

recstudio/data/config/amazon-books.yaml

@@ -23,7 +23,7 @@ field_separator: ","
 min_user_inter: 5
 min_item_inter: 5
 field_max_len: ~
-low_rating_threshold: ~
+low_rating_thres: ~
 max_seq_len: 20
 
 # network feature, including social network and knowledge graph, the first two fields are remapped the corresponding features

recstudio/data/config/amazon-electronics.yaml

@@ -23,7 +23,7 @@ field_separator: ","
 min_user_inter: 5
 min_item_inter: 5
 field_max_len: ~
-low_rating_threshold: 3
+low_rating_thres: 3
 max_seq_len: 20
 
 # network feature, including social network and knowledge graph, the first two fields are remapped the corresponding features

recstudio/data/config/coat.yaml

@@ -0,0 +1,47 @@
+url: https://rec.ustc.edu.cn/share/b097d230-c7e1-11ed-997a-bba57c4b600a
+user_id_field: &u user_id:token # TODO: comments for &u and *u
+item_id_field: &i item_id:token
+rating_field: &r rating:float
+time_field: ~ 
+time_format: ~
+
+
+inter_feat_name: coat.mnar
+inter_feat_field: [*u, *i, *r]
+inter_feat_header: 0
+
+user_feat_name: [coat.user]
+user_feat_field: [[*u, ugender(men):token, ugender(women):token, age(20-30):token, age(30-40):token, age(40-50):token, age(50-60):token, age(over 60):token, age(under 20):token, location(rural):token, location(suburban):token, location(urban):token, fashioninterest(moderately):token, fashioninterest(not at all):token, fashioninterest(very):token]]
+user_feat_header: 0
+
+
+item_feat_name: [coat.item]
+item_feat_field: [[*i, igender(men):token, igender(women):token, jackettype(bomber):token, jackettype(cropped):token, jackettype(field):token, jackettype(fleece):token, jackettype(insulated):token, jackettype(motorcycle):token, jackettype(other):token, jackettype(packable):token, jackettype(parkas):token, jackettype(pea):token, jackettype(rain):token, jackettype(shells):token, jackettype(track):token, jackettype(trench):token, jackettype(vests):token, jackettype(waterproof):token, color(beige):token, color(black):token, color(blue):token, color(brown):token, color(gray):token, color(green):token, color(multi):token, color(navy):token, color(olive):token, color(other):token, color(pink):token, color(purple):token, color(red):token, onfrontpage(yes):token, onfrontpage(no):token]]
+item_feat_header: 0
+
+
+field_separator: "\t"
+min_user_inter: 5
+min_item_inter: 5
+field_max_len: ~
+low_rating_thres: 3.0
+max_seq_len: 20
+
+# network feature, including social network and knowledge graph, the first two fields are remapped the corresponding features
+network_feat_name: ~ 
+mapped_feat_field: ~
+network_feat_field: ~
+network_feat_header: ~
+
+# interactions that are missing completely at random
+mcar_feat_name: coat.mcar
+mcar_feat_field: [*u, *i, *r]
+mcar_feat_header: 0
+
+# propensities of each (u, i) pair
+# propensity_feat_name: coat.propensities
+# propensity_feat_field: [*u, *i, propensity:float]
+# propensity_feat_header: 0
+
+
+save_cache: True # whether to save processed dataset to cache.

recstudio/data/config/gowalla.yaml

@@ -24,7 +24,7 @@ seq_separator: " "
 min_user_inter: 5
 min_item_inter: 5
 field_max_len: ~
-low_rating_threshold: ~
+low_rating_thres: ~
 max_seq_len: 20
 
 # network feature, including social network and knowledge graph, the first two fields are remapped the corresponding features

recstudio/data/config/ml-100k.yaml

@@ -21,8 +21,8 @@ item_feat_header: 0
 
 
 field_separator: "\t"
-min_user_inter: 0
-min_item_inter: 0
+min_user_inter: 5
+min_item_inter: 5
 field_max_len: ~
 low_rating_thres: 3.0
 max_seq_len: 20

recstudio/data/config/ml-10m.yaml

@@ -26,7 +26,7 @@ field_separator: "::"
 min_user_inter: 5
 min_item_inter: 5
 field_max_len: ~
-low_rating_threshold: 3.0
+low_rating_thres: 3.0
 max_seq_len: 20
 
 # network feature, including social network and knowledge graph, the first two fields are remapped the corresponding features

recstudio/data/config/ml-20m.yaml

@@ -26,7 +26,7 @@ field_separator: ","
 min_user_inter: 5
 min_item_inter: 5
 field_max_len: ~
-low_rating_threshold: ~
+low_rating_thres: ~
 max_seq_len: 20
 
 # network feature, including social network and knowledge graph, the first two fields are remapped the corresponding features