Have a separate function for the feature extraction

[relleums]

I propose to use a separate function to extract features from the photon_stream.
At the moment you have if else blocks for different treatment of simulation and observation events inside the feature extraction. This way, one might end up to have multiple of such if else blocks all over the code which is error prone.
I propose to make a separate function which takes only the photon_stream e.g. in point-cloud representation, and then returns a dictionary of the extracted features. The extraction of feature must be independent of the event being a simulation or observation, so I would show and emphasize this in the code.
This function is then called on both simulations and observations.
I would restrict my self to have only one of such if else blocks in the very beginning.

Also when the feature extraction is a function of its own, unit-tests are easier to implement (loosely coupled).

[KevSed]

So you propose to have one function to generate features from both, simulation and data? Currently there is a difference between these two when calculating features. There are for example simulation_truth that are saved from the MC (e.g. for training) but, of course, not available in data and observation_info that are saved from the observed data files. So using one function requires if else blocks anyway, no? Except there were two separate functions, which then would make it messy for the "shared" features to make changes...
I definitely see the danger of using too many if else blocks, though.

[relleums]

On the one hand side there is the representation of events (both simulations and observations) in the photon-stream and on the other side there is our intent to perform a certain task on certain parts of the data within these events.
The air-shower-features are independent of the simulation-truth. To express this, I would not pass on the event-class into the function which extracts the air-shower-features, but only the part of the event which is needed to extract the air-shower-features. This might be the raw-photon-stream, but not the full event. The feature-extraction does not need to access the ID of the event or its simulation-truth.
Imagine you want to set-up a unit-test. For a unit-test it is easier to create a fake raw-photon-stream than to also fake the rest of the event. This is about having loose couplings between the functions.
I think, I just recommend to split your processing into small functions which indicate their purpose already from their input and output arguments.