Tested the filter with the converter, larndsim and larnd2supera using two MR4 files (0533 and 0620). One of them (0533) has known issue (too many trajectories) at larcv stage (within larnd2supera).
With the filter, the converter stage is marginally faster, depending on the file. The converter run time w/ and w/o the filter on 0533 are 60s and 140s. The number of trajectories reduced from 686446 to 85122. The file size (0533) with filter is 30M while it's 100M w/o the filter. The converter run time w/ and w/o the filter on 0620 are 120s and 170s. The number of trajectories reduced from 500287 to 85174, roughly 6-8 times of shrinkage. The file size (0620) is 30M compared to 80M w/o the filter.
The vertex and segment datasets are unaffected.
Tested the filter with the converter, larndsim and larnd2supera using two MR4 files (0533 and 0620). One of them (0533) has known issue (too many trajectories) at larcv stage (within larnd2supera). With the filter, the converter stage is marginally faster, depending on the file. The converter run time w/ and w/o the filter on 0533 are 60s and 140s. The number of trajectories reduced from 686446 to 85122. The file size (0533) with filter is 30M while it's 100M w/o the filter. The converter run time w/ and w/o the filter on 0620 are 120s and 170s. The number of trajectories reduced from 500287 to 85174, roughly 6-8 times of shrinkage. The file size (0620) is 30M compared to 80M w/o the filter. The vertex and segment datasets are unaffected.