Add a DLD component

[philsmt]

For various delay line detectors in use at SQS, the offline calibration pipeline is used to digitize the analog signals recorded with digitizers and reconstruct them into proper detector hits in time and space as a virtual detector source. As this kind of data is variable-length by nature, it is saved in sufficiently large fixed-length arrays filled with np.nan and compressed away.

While this data format is very efficient for storage and interoperable, it is somewhat difficult to load and work on in a pulse or hit-based manner. Based on observations of existing user code, this component transform the train/pulse/hit-shaped data into a linear pd.Series or pd.DataFrame with multi indices for convenient access.

It integrates into the existing pulse pattern components, though by default will use the internal pulse information saved as part of the processing result.

[philsmt]

Reminder to myself: Add pulse height information to edges, if available.

[philsmt]

@JamesWrigley Could I ask you as well to have a brief look? It's been dangling for a bit now, and I would like to base more components off it.

[JamesWrigley]

I should say that I'm completely unfamiliar with DLD's, so I may have missed something 👀

[JamesWrigley]

LGTM!

[takluyver]

I found the _get_reduced_pd method not especially easy to read. It feels like it's trying to cover a lot of possibilities for something that's only called in 3 places. I suspect that it could be broken up a bit, but I haven't understood it well enough to suggest how. That's not really important if you're happy with it, though.

I think my question about RUN values is worth thinking about before we merge this.

[philsmt]

Do you feel strongly about the complexity of _get_reduced_pd? I also dislike the split_axis part, but ultimately it made the getter methods shorter. For now it's a private method anyway, though I could imagine it being more generally useful at some point.
I experimented with splitting it up in two methods, one for building an aligned index and one to reduce sparse datasets into dataframes, but it ended up clogging the getter methods...

[takluyver]

No, I don't feel strongly about it. 🙂

[philsmt]

@takluyver I ended up giving it another try to reduce the complexity of _get_reduced_pd by splitting it along the messy "it-may-be-a-list-or-not" line. Is this more readable and/or understandable?

[takluyver]

Thanks, that does look easier to read now.

[philsmt]

Thanks for review!