Documentation on modifying events table

Since it is not possible to rewrite Observation.events, we should show some documentation on how to modify the events list.

1. The minimal solution would be to provide a HowTo entry with maybe the following lines of code

import numpy as np
from gammapy.data import DataStore, Observation, EventList

datastore = DataStore.from_dir("$GAMMAPY_DATA/hess-dl3-dr1/")
obs = datastore.obs(23523)

# compute the phase separately, using, eg: PINT
phase = np.random.random(len(obs.events.table))

# attach the phase column to the events table
table = obs.events.table
table["PHASE"] = phase
events_new = EventList(table)

# create a new observation
irfs = {}
for _ in obs.available_irfs:
    irfs[_]= getattr(obs,_)
obs_new = = Observation.create(pointing=obs.pointing_radec, 
                  location=obs.observatory_earth_location,
                  obs_id=obs.obs_id,
                  tstart=obs.tstart,
                  tstop=obs.tstop,
                  irfs=irfs)

1. We can modify Observation.create() or add a Observation.copy() to simplify this. One of the problems with the above lines of code is that most of the information contained in obs.fixed_pointing_info.meta is lost in obs_new

I can implement either. Preference @registerrier @maxnoe @adonath ?

[maxnoe]

One point regarding pulsar phase: DL3 files are source agnotic and phase is a property of an assumed source. There could even be the case where you have two pulsars in the same FoV, which will probably be the case for CTA observations of the galactic plane?

So having a single phase column in dl3 event lists does not really match the data model. You’d need a phase column per assumed object.

Of course, in most current cases, there is only one.

But at least CTA will very probably not have phase information prefilled in DL3 files, since the only relevant information from CTA is the timestamp of the events and the rest has to come from external information about the assumed object(s)…

[adonath]

I think having a convenience method to load data from an observation into memory might be useful here. So either:

• Observation.to_in_memory() or similar
• data_store.obs(12346, in_memory=True)