Permalink
Find file
Fetching contributors…
Cannot retrieve contributors at this time
40 lines (32 sloc) 2.71 KB
VERY NICE....We started doing Affine System VERY VERY wrong.
(I didn't want to look up the damn algorithm...still haven't but after walking it through on
the whiteboard its lear we need a system apart from the NWMatrix.
Specifically the matrices D, and I (both the same size as NWMatrix (M))
We'll keep the Affine Automata System but we'll use it here to actually fill both D and I before
attacking the NWMatrix.
With this we can simplify the appearence saying
setMatrixValue(_system._matrix, i, j) = max( F_imin1_jmin1 + s(charati,charatj),
F_i_jmin1 - D(i,j) #note here that D,I r level to M
F_imin1_j - I(i,j)
)
Mind you I may be STILL confused about insertions/deletions being wrt which of the two sequences, I think
its the sequence on top along j, which for some reason I think I refer to often as the "second sequence"
=> even though I think of the top sequence (along j) as being the "first sequence"
(the point is I don't think its healthy to be confused and having semantics that to me are
counter intuitive -- fix all this shit please todo:will March 9, 2013)
Note: This is all is very much like the part we have to do for the recognition,
so while we're here try and do it right...like use 2D arrays or something
OOOOH, and I should mention we sized the NWMatrix wrong...it should be ((length(S1)+1) x (length(S2)+1))
the real processing should of course then still be restricted to i = [2, |S1|+1] and j = [2, |S2|+1]
By the way...githubs diff isn't that great, just realized it with the last commit (interesting...)
(todo:will see notes in blue book on this)
From our mistakes /[(|lets hope)] we become stronger and less stupid
March 9, 2013 - 2:37 Note:
Okay, so this is good! Those seperate matrices should(I think must, but not gonna go into that) be evaluated alongside the NWMatrix
Oop, also. The NW Matrix is "wrong". Unless we WANT to allow gaps at the ends...there's some theory on this I think,
but I think generally we dont(?) with "global alignments"(?) >>Check it out<<
Just want to note this pass isn't "optimal". In particular we're using Dictionaries accessible by tuples for the NW Matrix,
this is only because we didn't look around long enough to find that there is a 2D Array (I think its fairly low level)
that we should (I think) use instead. (Laziness...)
- March 8, 2013
We need to scrap all the output calls once thing are near finalized to set up profiling.