Records of Unequal Size

[joefowler]

Original report by Galen O'Neil (Bitbucket: oneilg, GitHub: oneilg).

---

This is is an outline of the work necessary to implement Records of Unequal Size (ROUS). There are a variety of ways to achieve this, I’m going to outline the way that seems to require the least research to implement. The idea is to generate N sets of M projectors, where each set has Ri samples and Pi pretrigger samples. For a particular record, one will select the projector set with the largest Pi from among the available projector sets, then truncate the record if neccesary to match the Ri and Pi of the projectors set, then with that set to generate the modeled pulse reduced MPR coefficients.

‌

Necessary steps (in order how I think I would approach them):

1. make_projectors must be be able to make a projectors set
2. Dastard must have an API to set a projectors set
3. Dastard Commander must expose an interface to load and send a projector set
4. Dastard must implement logic to choose projectors from the set and generate coefs with them
5. The OFF file format must change to include the full projectors set
6. dastard needs to write the new OFF format
7. mass.off needs to load the new format
8. Mass.off should have some sensible default for dealing with ROUSs, I propose the following: change the default cut from cutNone to cutShort where cutShort marks all records with less than the maximum length as bad. This will likely evolve over time, but it is the closest to what we currently do, and will make all new stuff opt-in.

Notes:

1. ljh2off needs to be able to read whatever make_projectors makes, but can’t handle variable length records until LJH handles variable length records
2. @{557058:23f7bf33-90da-4f11-9045-26292ebeefb8} has a way he thinks is better where you make some sort of noise model with special properties that lets you somehow only have one projector set and truncate it as needed. But we dont have the code (and the progress that exists is in Julia) to actually make the needed parts. Looking at this list, only 8 seems completely re-usable between both approaches, which is a bit sad.

‌

[joefowler]

Original comment by Galen O'Neil (Bitbucket: oneilg, GitHub: oneilg).

---

We decided not to work on this immediatley, its too much work, with too little re-usable for what we would really like to do.

[joefowler]

Original comment by Joseph Fowler (Bitbucket: joe_fowler, ).

---

Removing milestone: v0.7 (automated comment)

[joefowler]

Original comment by Joseph Fowler (Bitbucket: joe_fowler, ).

---

No progress in 3 years, and not at all clear what to do.