Skip to content

Latest commit

 

History

History
68 lines (50 loc) · 2.46 KB

Hacks.md

File metadata and controls

68 lines (50 loc) · 2.46 KB
Raw

Something something hacks

These are a list of small projects that can be tackled on Tuesday and Wednesday. The goal is to gain some experience with each topic, share this experience with others, and build models that will make it easier to discuss each idea in a concrete way.

abstraction levels

The higher level abstraction we can offer the "user" the more opportunity we have for optimising things. What is the right level of abstraction? Use cases from fitting tracks, doing things to a group of particles, selecting a range of events to store in a file.

Random access store

What technology exists out there for efficient storage that fits our needs and allows for random access. A small review of the existing chunked, column stores could be a good starting point. The advantage of being able to do random access would be not having to read through large numbers of files that are in the same stripping stream.

storing objects

Right now it is impossible to add a new class to be stored in a DIGI/DST. This leads to all sorts of hacks. boost::any? captnproto? What are possible answers? This is related to backwards compatibility in #12. Keyword: type erasure. The IO layer should be contained in a small, standalone library which will make it more portable/reusable. Think toolkit instead of framework.

linking information across objects

We invented ExtraInfo as some crazy hack that does not work very well. We need a better way to be able to add more information to a given object (be that a track, particle, hit, ...). This information will exist for some but not all objects in a collection. If you think of the information in an event as a big table, then this is natural, you simply add a new column. Think of composing more complex objects from simpler ones.

best way to express a processing pipeline

How do you express something like "take these hits, build tracks from them, filter on X, take these other hits, build different kinds of tracks from them, filter them on Y, merge the two track collections, evaluate PID info for them, select pions with pt>5GeV" in a way that is flexible and extremely easy for humans to read, and also possible for computer to parse? The aim is to write this kind of processing using ideas from functional programming:

even = filter(lambda x: x%2, range(10))
even_squared = map(lambda x: x**2, even)

# comapred to
def make_even_squares():
  out = []
  for n in range(10):
    if n%2:
      out.append(n**2)
  return out