In Computational-Cognitive-Musicology-Lab/humdrumR: humdrumR

source('vignette_header.R')

Welcome to "Grouping humdrum data"! This article explains functionality r hm has to break your data into subgroups, and work with each subgroup separately.

In this article, we'll once again work with our prepackaged chorales dataset, as we did in our quick start article.

chorales <- readHumdrum(humdrumRroot, 'HumdrumData/BachChorales/chor.*.krn')

chorales

Group-apply-combine

One of the fundamental data analysis techniques in R is the group-apply-combine routine. We break our data.frames into groups (by row), apply calculations or manipulations to each group independently, and then (optionally) recombine the groups back into a single data.frame. For example, if we want know the average pitch height (in MIDI pitch) of the chorales, we could run this command:

chorales |>
  midi() |>
  summarize(mean(.))

(We use the summarize() command because we are only getting one (scalar) value.) But what if wanted to calculate the mean pitch, for each chorale? We can use the group_by() command:

chorales |>
  midi() |>
  group_by(Piece) |>
  summarize(mean(.))

How does this work? The group_by() function looks at all the unique values the fields we give it---in this case the Piece field---and breaks the humdrum table into groups based on those values. So where ever Piece == 1, that's a group; wherever Piece == 2, that's another group, etc. After group_by() is applied, the r hm dataset is grouped, and any subsequent tidy-commands will automatically applied within the groups.

---

Once a r hm dataset has been grouped, it will stay grouped until you call ungroup() on it!

More complex grouping

If we like, we can pass more than one field to group_by(). Groups will be formed for every unique combinations of values across grouping fields:

chorales |>
  midi() |>
  group_by(Piece, Spine) |>
  summarize(mean(.))

You can also create new grouping fields on the fly.F For example, what if, instead of groups for four separate voices, we wanted to group bass/tenor and alto/soprano within each piece? We could do this:

chorales |>
  midi() |>
  group_by(Piece, Spine > 2) |>
  summarize(mean(.))

---

Grouping isn't just good for calcualting single values. We can, for instance, tabulate each group separately:

chorales |>
  kern(simple = TRUE) |>
  group_by(Spine) |>
  count() 

Common groupings

By file

It is very common in r hm analyses that we want to apply commands grouped by piece. In most cases, two pieces of music are completely separate entities, so it makes sense to apply calcualtions/manipulations separately to each piece. However, r hm won't (always) do this for you. You might think, "I want to count how many notes occur in each record" and run the command group_by(Record)---but watch out! Grouping by Record will group all the records 1s across all the pieces, all the record 2s across all the pieces, etc. What you probably want is (group_by(Piece, Record)). Don't forget to group by piece...unless you are sure you don't want to.

Melodic grouping

In lots of humdrum data, each spine (or at least some spines) in the data represent a musical part of voice. We often want isolate each voice, so group_by(Piece, Spine) gets used a lot. However, watch out for spine paths and multi-stops, if your data has them---you might want something like group_by(Piece, Spine, Path), for example, if you data has spine paths. Check out the [complex syntax][ComplexSyntax.html "Complex humdrum syntax article"] for more ideas/options about this.

Recycling results

As usual, if you use mutate() with grouped data, the results of you command will be "recycled"---in this case, to fill each group. This can be very useful for maintaining vectorization. For example, let's say we want to calculate the lowest note in each bar.

chorales |>
  semits() |>
  group_by(File, Bar) |>
  mutate(BarBassNote = min(Semits)) |>
  ungroup() -> chorales

chorales

What did this command do? It looked at every bar in each file and found the lowest note, then "recycled" that note throughout the bar. Since the lowset-notes have been recycled, there is one value for each and every note in Token so everything is still neatly vectorized. This means we can calculate the harmonic interval above each bar's bass note by just going:

chorales |>
  mutate(Semits - BarBassNote) |>
  pull() |>
  draw()

Computational-Cognitive-Musicology-Lab/humdrumR documentation built on Oct. 22, 2024, 9:28 a.m.