Contour & height-based gaps/mismatches highlighting

[notkaramel]

The current implementation of the analysis tool using Needleman-Wunsch algorithm is highlighting the contour between two chants. These mismatches and gaps are made of two notes, and in order to highlight them, we must highlight both notes that create it. This is a bit unintuitive and might be confusing to visually analyse.

To-do (for cross-comparison):

• [ ] Implement a location-based analysis
  • [ ] Rebasing both chant to start at location 0.
• [ ] Highlight location-based analysis
• [ ] ~Implement and test a different algorithm (e.g., Smith-Waterman) for cross-comparison analysis~
• [x] Improve the current contour highlight feature to be more intuitive
• [ ] Allowing to highlight both mismatches and gaps on both sides by adding an extra radio button that does mismatches with spotlight highlighting: https://github.com/ECHOES-from-the-Past/mei-analyser/commit/2ec4c399fc72142b1ff6783d4892d59e4cfe4707
• [ ] (For contour-base cross-comparison, only highlight one of the two notes that make up the melodic interval: https://github.com/ECHOES-from-the-Past/mei-analyser/commit/eed5a065f06c8666477aa094e92825faeaa5d0ec)
• [ ] Account for gap + mismatch edge cases (chant 05-06)
• [ ] Makes highlighting easier to interpret - one note only, two or more might be confusing for the user.
• [ ] Extra developer tool (will not be in final interface): selection for gaps, match, mismatch score
• [ ] Extra: Testing the implementation of penalty matrix

[martha-thomae]

I am including here a few afterthoughts I had:

• The actions of "substituting a mismatch for the correct note" and "deleting an extra note" (or, equivalently, "adding a missing note") have an effect that propagates to the next element of the melodic-interval sequence (or contour sequence). For example: if we substitute a mismatched note, we are changing the melodic interval of that note with the previous note (let's call it the "current interval"), but we are also changing the coming melodic interval (the one formed with the next note). So far, we are not considering this effect on the next element of the contour sequence when we make any time of edit (substitution of mismatch, add extra note, or delete extra note). Therefore, if we keep working with sequences of melodic intervals, we need to consider the effect of the edit operation (substitution, addition, deletion of note) in the following interval.
• The reason why I suggested working with sequences of melodic intervals (or contour sequences) rather than with note-heights sequences in the first place was because a contour sequence does not need to consider transposition (transporting the element up or down by a fixed interval). For example: d d f will be equal to g g b if we consider the sequence of intervals 0 +2, but they have different "heights" (different pitches). So I was thinking that something like: c d d f vs. e g g b, will have the contour +1 0 +2 vs. +2 0 +2 and, therefore, the difference only lies on the first note. If we are going to switch now to a "note-height" approach, we will need to consider an extra edit operation (in addition to the usual substitution, addition, and deletion): transposition, which in this context might mean adding or subtracting a fixed interval to a sub-sequence of notes. So maybe we should start looking for melodic comparison (with Needleman-Wunsch or others) that include a transposition operation.

[notkaramel]

Thank you @martha-thomae, that's very insightful!

I've been debugging the current Needleman-Wunsch algorithm for the past couple of days and see if it can construct and analyse chants via its height/position first.

"[...] if we keep working with sequences of melodic intervals, we need to consider the effect of the edit operation (substitution, addition, deletion of note) in the following interval."

But isn't it the goal of the algorithm is to highlight these edits while keeping the the remainder of the chant to stay the same?

"[...] If we are going to switch now to a "note-height" approach, we will need to consider an extra edit operation (in addition to the usual substitution, addition, and deletion): transposition, which in this context might mean adding or subtracting a fixed interval to a sub-sequence of notes."

I agree and am testing this approach. Will update soon!

Sidenote: I made some indexing mistake for the matrix, making the row to be the column and vice versa. This lead to the analysis to be swapped. It's only recently that I fixed this issue.

[notkaramel]

• I will start with this feature since #5 is now fixed :)

[martha-thomae]

To-do (for cross-comparison):

• [ ] Aquitanian: based on @loc
• [ ] Square: setting the first note to be location 0, and subsequence note are relative to it.
• [ ] Highlight location-based analysis
• [x] Improve the current contour highlight feature to be more intuitive
• [ ] Allowing to highlight both "mismatches" and "extra notes" on both sides by adding an extra radio button for "All": https://github.com/ECHOES-from-the-Past/mei-analyser/commit/2ec4c399fc72142b1ff6783d4892d59e4cfe4707. (Please do mismatches with spotlight highlighting.)
• [ ] Also use red font for the integer corresponding to the extra note: https://github.com/ECHOES-from-the-Past/mei-analyser/commit/64943ed41ba28585f978a27412ae8b2b2d0f172b.
• [ ] (For contour-base cross-comparison, only highlight one of the two notes that make up the melodic interval: https://github.com/ECHOES-from-the-Past/mei-analyser/commit/eed5a065f06c8666477aa094e92825faeaa5d0ec)

I removed the following item from the to-do list:

• [ ] Implement and test a different algorithm (e.g., Smith-Waterman) for cross-comparison analysis

[notkaramel]

Adding extra requirements and ideas/concerns (discussed with @martha-thomae) about highlight mismatches and gaps for cross comparison:

• For height-based comparison (high priority):
  • [ ] Rebase the first note of both chants (Aquitanian and Square) to 0
• [ ] Account for gap + mismatch edge cases (chant 05-06)
• [ ] Makes highlighting easier to interpret - one note only, two or more might be confusing for the user.
• [ ] Extra developer tool (will not be in final interface): selection for gaps, match, mismatch score
• [ ] Extra: Testing the implementation of penalty matrix

[notkaramel]

Delaying implementation of this feature (again) to focus on search functionalities

[martha-thomae]

location + show differences on both sides