use_id 0 in generate_data() actually has two records with the same timestep 300, which are the last interactions for use_id 0. When we do the leave-one-out split, there will be two versions of splits that are both valid. But current testing codes only takes the row index as an indication for the asserting, which will only accept one splitting method.
Describe the bug
with the following code,
use_id 0 in generate_data() actually has two records with the same timestep 300, which are the last interactions for use_id 0. When we do the leave-one-out split, there will be two versions of splits that are both valid. But current testing codes only takes the row index as an indication for the asserting, which will only accept one splitting method.
https://github.com/beta-team/beta-recsys/blob/91742c25d38c0c1a9008cec5e77a9a94df9bf16b/tests/test_data_split.py#L245
To Reproduce
See the example blew, which is a valid split, but was rejected by the test.