Currently, the log split is performed by iterating trace by trace, adding it to split_1, until the number of events in split_1 is enough given the specified partition ratio (e.g., 80%-20%).
This process could be done following an approach that selects the % of cases matching the partition ratio, checks whether this is a good event-wise partition, and if not, adjusts the selected partition by adding/subtracting half of the remaining cases, if not enough, again adjust by half. In this way, the operation should converge faster, especially for event logs with many cases (like CallCentre), which currently are taking too long for this.
Currently, the log split is performed by iterating trace by trace, adding it to
split_1, until the number of events insplit_1is enough given the specified partition ratio (e.g., 80%-20%).This process could be done following an approach that selects the % of cases matching the partition ratio, checks whether this is a good event-wise partition, and if not, adjusts the selected partition by adding/subtracting half of the remaining cases, if not enough, again adjust by half. In this way, the operation should converge faster, especially for event logs with many cases (like CallCentre), which currently are taking too long for this.