Already in the very old versions of protopipe, it was possible to select a cleaning which retained the number of islands.
This has never been used (at least based on the common tailcut cleaning) and the main cluster cleaning has been chosen for both,
image cleaning of events for DL2 production,
image cleaning for training
which is also what (at least) CTAMARS does.
EDIT: this is not correct, see issue #117 - still important to check the effect of clusters on classifier training!
In case the image is produced by a proton (electrons tend to produce gamma-like images due to the nature of EM showers) it is expected (but never tested in protopipe) that the number of surviving clusters could be useful for classification purposes (by becoming a feature in the classification models).
This shouldn't be instead important for energy training, since there we already use images exclusively produced by gammas.
Already in the very old versions of protopipe, it was possible to select a cleaning which retained the number of islands.
This has never been used (at least based on the common tailcut cleaning) and the main cluster cleaning has been chosen for both,
which is also what (at least) CTAMARS does.
EDIT: this is not correct, see issue #117 - still important to check the effect of clusters on classifier training!
In case the image is produced by a proton (electrons tend to produce gamma-like images due to the nature of EM showers) it is expected (but never tested in protopipe) that the number of surviving clusters could be useful for classification purposes (by becoming a feature in the classification models).
This shouldn't be instead important for energy training, since there we already use images exclusively produced by gammas.