使用recbole1.2.0时发现ml-1m的数据数量对不上

[yunshanlucky]

模型用的是kgat，数据集选用ml-1m，发现无论是原始未作任何处理的ml-1m数据集，还是官方提供的最优的数据配置下的用户数项目数，都与我操作出来的实际数量对不上，

以下是我复现未作任何处理的数据集，其中官方给出的用户数为6041，物品数量为3707，结果不一致
17 May 10:42 INFO ml-1m
The number of users: 6041
Average actions of users: 165.16225165562915
The number of items: 3656
Average actions of items: 272.93570451436386
The number of inters: 997580
The sparsity of the dataset: 95.48318075934071%
Remain Fields: ['entity_id', 'user_id', 'item_id', 'rating', 'head_id', 'relation_id', 'tail_id']
The number of entities: 79348
The number of relations: 51
The number of triples: 385923
The number of items that have been linked to KG: 3655

同时给出我复现的官方最优参数的结果，其中官方给出的用户数为6034，物品数量为3096，结果不一致
17 May 11:19 INFO ml-1m
The number of users: 6034
Average actions of users: 137.95773247140727
The number of items: 3104
Average actions of items: 268.2239767966484
The number of inters: 832299
The sparsity of the dataset: 95.55622200144201%
Remain Fields: ['entity_id', 'user_id', 'item_id', 'rating', 'head_id', 'relation_id', 'tail_id']
The number of entities: 17012
The number of relations: 100
The number of triples: 483000
The number of items that have been linked to KG: 3103

以下是官方中找到的最佳配置信息：

dataset config

field_separator: "\t"
seq_separator: " "
USER_ID_FIELD: user_id
ITEM_ID_FIELD: item_id
RATING_FIELD: rating
HEAD_ENTITY_ID_FIELD: head_id
TAIL_ENTITY_ID_FIELD: tail_id
RELATION_ID_FIELD: relation_id
ENTITY_ID_FIELD: entity_id
NEG_PREFIX: neg_
LABEL_FIELD: label
load_col:
inter: [user_id, item_id, rating]
kg: [head_id, relation_id, tail_id]
link: [item_id, entity_id]

data filtering for interactions

val_interval:
rating: "[3,inf)"
unused_col:
inter: [rating]

user_inter_num_interval: "[10,inf)"
item_inter_num_interval: "[10,inf)"

data preprocessing for knowledge graph triples

kg_reverse_r: True
entity_kg_num_interval: "[5,inf)"
relation_kg_num_interval: "[5,inf)"

training and evaluation

epochs: 500
train_batch_size: 4096
eval_batch_size: 40960000
valid_metric: NDCG@10
train_neg_sample_args:
distribution: uniform
sample_num: 1
dynamic: False

model

embedding_size: 64