The current version of the split method can accidentally cut off some users and interactions from the dataset. It happens when the split point occurs at the same time a new user enters the system and is designated to the train dataset. For example, the timestamp threshold is 500 and a train user first interaction is 500 as well. By the current condition, it wouldn't be in neither train dataset nor test dataset. This PR fix this situation.
The current version of the split method can accidentally cut off some users and interactions from the dataset. It happens when the split point occurs at the same time a new user enters the system and is designated to the train dataset. For example, the timestamp threshold is 500 and a train user first interaction is 500 as well. By the current condition, it wouldn't be in neither train dataset nor test dataset. This PR fix this situation.