Regex refactor

[trevorsummerssmith]

• Figure out what to do about the triple (graph, transition, transition_op)
  should we have a class for this? Is it a thing or do we need to refactor
  the regex stuff a bit?
• Rename the 'filter.py' module and its classes. It's confusing to have a filteregex
  and the filter paradigm. Maybe:
  • pattern is a regex
  • sensor is a ball
  • combinator is a stick
  • or
  • sensor operator = ball
  • nexus operator = stick
  • branch operator = rank-3 and higher tensors

Also make a 'regex' module. 'regex_parser' => ast. Make things more sane.

[epurdy]

I vote for

• pattern (was: regex)
• sensor (was: filterregex)
• nexus (was: graph, transition, transition_op)
• combinator (rank-3 and higher tensors)

[trevorsummerssmith]

I think I'll add these changes. However, I want to keep the comments that talk about regexes (eg "We wish to manipulate probability distributions over the nodes of a Markov chain. ... Regexes are an intuitive way to specify such linear operators.").

But then just add some stuff that says, at the end of the regex.py module comment "However note that we wish to distinguish between three types of regexes: sensor, nexus and combinator".

Thoughts?

[trevorsummerssmith]

Actually I think I like 'regex' better than 'pattern'. More thoughts? @epurdy

[epurdy]

"regex" does sound better than "pattern", but at some point we should give it a new name, just so that it doesn't sound like a classical regular expression. I think they're strictly more powerful than regular expressions, and if they actually become common, we're going to want to distinguish between them.

I'm having trouble thinking of what to call it, though. "nexop" (nexus operator)? "semex" (semiotic expression)? "semop" (semiotic operator)? "gramex" (grammatical expression)? "gramop" (grammatical operator)?

If I had to choose any of those, I think I'd choose semex.