make lt-comp go a bit faster #114
Conversation
There was a problem hiding this comment.
mind filling in the git commit messages a bit with the reasoning? (the current message just makes it seem like you fixed a missing thing in the dtd and made some internal speed optimization, but IIUC it's actually a new feature that you have to apply to your dix files to make them compile faster)
| paradigms[current_paradigm].minimize(); | ||
| } else { | ||
| paradigms[current_paradigm].joinFinals(); | ||
| } |
There was a problem hiding this comment.
Shouldn't this be if(current_minimise) minimize(); joinFinals() , ie. we want to joinFinals also when we minimize? (Seems like the current change would also alter how non-paradigms are compiled!)
There was a problem hiding this comment.
minimize() calls reverse() which calls joinFinals(), so there shouldn't be a change in behavior.
There was a problem hiding this comment.
Calls reverse twice in fact, so it both joins finals and initials … I suppose nowhere in that sequence any new (unjoined) final is added so it's safe then :)
| for(auto& it2 : finals) { | ||
| finals_state.insert(it2.first); | ||
| } | ||
|
|
There was a problem hiding this comment.
for our edification, can you explain why this block is moved?
There was a problem hiding this comment.
It's calculating a list of the final states in the original transducer, but in its original location it was recalculating it on each iteration but it won't have changed.
|
LGTM |
Compiling the cat monodix was taking about 15 seconds that seemed kind of high to me so I implemented the changes mentioned in #96 and also poked at some other things and managed to get the cat compile time down to about 12.5 seconds.
From profiling it looks like 80% of the runtime of
lt-compis intransducer.minimize()which is inconveniently also the the one step of the process that I don't actually understand.