Annotations#

Hide imports 
%load_ext autoreload
%autoreload 2

import os

import dimcat as dc
import ms3
import plotly.express as px
from dimcat import groupers, plotting

import utils
Hide source 
RESULTS_PATH = os.path.abspath(os.path.join(utils.OUTPUT_FOLDER, "overview"))
os.makedirs(RESULTS_PATH, exist_ok=True)


def make_output_path(
    filename: str,
    extension=None,
    path=RESULTS_PATH,
) -> str:
    return utils.make_output_path(filename=filename, extension=extension, path=path)


def save_figure_as(
    fig, filename, formats=("png", "pdf"), directory=RESULTS_PATH, **kwargs
):
    if formats is not None:
        for fmt in formats:
            plotting.write_image(fig, filename, directory, format=fmt, **kwargs)
    else:
        plotting.write_image(fig, filename, directory, **kwargs)

Loading data

Hide source 
D = utils.get_dataset("beethoven_piano_sonatas", corpus_release="v2.5")
package = D.inputs.get_package()
package_info = package._package.custom
git_tag = package_info.get("git_tag")
utils.print_heading("Data and software versions")
print("Ludwig van Beethoven – Piano Sonatas version v2.5")
print(f"Datapackage '{package.package_name}' @ {git_tag}")
print(f"dimcat version {dc.__version__}\n")
D

Data and software versions
--------------------------

Ludwig van Beethoven – Piano Sonatas version v2.5
Datapackage 'beethoven_piano_sonatas' @ v2.5
dimcat version 3.4.0

Dataset
=======
{'inputs': {'basepath': None,
            'packages': {'beethoven_piano_sonatas': ["'beethoven_piano_sonatas.measures' "
                                                     '(MuseScoreFacetName.MuseScoreMeasures)',
                                                     "'beethoven_piano_sonatas.notes' "
                                                     '(MuseScoreFacetName.MuseScoreNotes)',
                                                     "'beethoven_piano_sonatas.expanded' "
                                                     '(MuseScoreFacetName.MuseScoreHarmonies)',
                                                     "'beethoven_piano_sonatas.chords' "
                                                     '(MuseScoreFacetName.MuseScoreChords)',
                                                     "'beethoven_piano_sonatas.metadata' (FeatureName.Metadata)"]}},
 'outputs': {'basepath': None, 'packages': {}},
 'pipeline': []}

filtered_D = D.apply_step("HasHarmonyLabelsFilter")
all_metadata = filtered_D.get_metadata()

assert len(all_metadata) > 0, "No pieces selected for analysis."
chronological_corpus_names = all_metadata.get_corpus_names()

DCML harmony labels#

Hide source 
all_annotations = filtered_D.get_feature("DcmlAnnotations")
is_annotated_mask = all_metadata.label_count > 0
is_annotated_index = all_metadata.index[is_annotated_mask]
annotated_notes = filtered_D.get_feature("notes").subselect(is_annotated_index)
print(f"The annotated pieces have {len(annotated_notes)} notes.")

The annotated pieces have 167022 notes.

all_chords = filtered_D.get_feature("harmonylabels")
print(
    f"{len(all_annotations)} annotations, of which {len(all_chords)} are harmony labels."
)

21788 annotations, of which 21509 are harmony labels.

Harmony labels#

Unigrams#

For computing unigram statistics, the tokens need to be grouped by their occurrence within a major or a minor key because this changes their meaning. To that aim, the annotated corpus needs to be sliced into contiguous localkey segments which are then grouped into a major (is_minor=False) and a minor group.

root_durations = (
    all_chords[all_chords.root.between(-5, 6)]
    .groupby(["root", "chord_type"])
    .duration_qb.sum()
)
# sort by stacked bar length:
# root_durations = root_durations.sort_values(key=lambda S: S.index.get_level_values(0).map(S.groupby(level=0).sum()),
# ascending=False)
bar_data = root_durations.reset_index()
bar_data.root = bar_data.root.map(ms3.fifths2iv)
fig = px.bar(
    bar_data,
    x="root",
    y="duration_qb",
    color="chord_type",
    title="Distribution of chord types over chord roots",
    labels=dict(
        root="Chord root expressed as interval above the local (or secondary) tonic",
        duration_qb="duration in quarter notes",
        chord_type="chord type",
    ),
)
fig.update_layout(**utils.STD_LAYOUT)
save_figure_as(fig, "chord_type_distribution_over_scale_degrees_absolute_stacked_bars")
fig.show()

relative_roots = all_chords[
    ["numeral", "duration_qb", "relativeroot", "localkey_is_minor", "chord_type"]
].copy()
relative_roots["relativeroot_resolved"] = ms3.transform(
    relative_roots, ms3.resolve_relative_keys, ["relativeroot", "localkey_is_minor"]
)
has_rel = relative_roots.relativeroot_resolved.notna()
relative_roots.loc[has_rel, "localkey_is_minor"] = relative_roots.loc[
    has_rel, "relativeroot_resolved"
].str.islower()
relative_roots["root"] = ms3.transform(
    relative_roots, ms3.roman_numeral2fifths, ["numeral", "localkey_is_minor"]
)
chord_type_frequency = all_chords.chord_type.value_counts()
replace_rare = ms3.map_dict(
    {t: "other" for t in chord_type_frequency[chord_type_frequency < 500].index}
)
relative_roots["type_reduced"] = relative_roots.chord_type.map(replace_rare)
# is_special = relative_roots.chord_type.isin(('It', 'Ger', 'Fr'))
# relative_roots.loc[is_special, 'root'] = -4

root_durations = (
    relative_roots.groupby(["root", "type_reduced"])
    .duration_qb.sum()
    .sort_values(ascending=False)
)
bar_data = root_durations.reset_index()
bar_data.root = bar_data.root.map(ms3.fifths2iv)
root_order = (
    bar_data.groupby("root")
    .duration_qb.sum()
    .sort_values(ascending=False)
    .index.to_list()
)
fig = px.bar(
    bar_data,
    x="root",
    y="duration_qb",
    color="type_reduced",
    barmode="group",
    log_y=True,
    color_discrete_map=utils.TYPE_COLORS,
    category_orders=dict(
        root=root_order,
        type_reduced=relative_roots.type_reduced.value_counts().index.to_list(),
    ),
    labels=dict(
        root="intervallic difference between chord root to the local or secondary tonic",
        duration_qb="duration in quarter notes",
        type_reduced="chord type",
    ),
    width=1000,
    height=400,
)
fig.update_layout(
    **utils.STD_LAYOUT,
    legend=dict(
        orientation="h",
        xanchor="right",
        x=1,
        y=1,
    ),
)
save_figure_as(fig, "chord_type_distribution_over_scale_degrees_absolute_grouped_bars")
fig.show()

print(
    f"Reduced to {len(set(bar_data.iloc[:,:2].itertuples(index=False, name=None)))} types. "
    f"Paper cites the sum of types in major and types in minor (see below), treating them as distinct."
)

Reduced to 57 types. Paper cites the sum of types in major and types in minor (see below), treating them as distinct.

dim_or_aug = bar_data[
    bar_data.root.str.startswith("a") | bar_data.root.str.startswith("d")
].duration_qb.sum()
complete = bar_data.duration_qb.sum()
print(
    f"On diminished or augmented scale degrees: {dim_or_aug} / {complete} = {dim_or_aug / complete}"
)

On diminished or augmented scale degrees: 70.41666666666667 / 35663.125 = 0.0019744951309417407

chords_by_mode = groupers.ModeGrouper().process(all_chords)
chords_by_mode.format = "scale_degree"

Whole dataset#

unigram_proportions = chords_by_mode.get_default_analysis()
unigram_proportions.make_ranking_table()
mode major minor
chord_and_mode scale_degrees duration_qb proportion proportion_% chord_and_mode scale_degrees duration_qb proportion proportion_%
rank
1 I, major (1, 3, 5) 4589.083333 0.193962 19.4 % i, minor (1, 3, 5) 2007.333333 0.167231 16.72 %
2 V7, major (5, 7, 2, 4) 2106.750000 0.089044 8.9 % V, minor (5, #7, 2) 1157.875000 0.096462 9.65 %
3 V, major (5, 7, 2) 2081.250000 0.087966 8.8 % V7, minor (5, #7, 2, 4) 970.750000 0.080873 8.09 %
4 I6, major (3, 5, 1) 1362.625000 0.057593 5.76 % i6, minor (3, 5, 1) 536.375000 0.044685 4.47 %
5 V(64), major (5, 1, 3) 744.333333 0.031460 3.15 % V(64), minor (5, 1, 3) 516.000000 0.042988 4.3 %
... ... ... ... ... ... ... ... ... ... ...
835 vi7(11), major (6, 1, 3, 5) 0.250000 0.000011 0.0 % NaN NaN NaN NaN NaN
836 ii43(4), major (6, 1, 2, 5) 0.250000 0.000011 0.0 % NaN NaN NaN NaN NaN
837 vi7(4), major (6, 2, 3, 5) 0.250000 0.000011 0.0 % NaN NaN NaN NaN NaN
838 V7(64), major (5, 1, 3, 4) 0.250000 0.000011 0.0 % NaN NaN NaN NaN NaN
839 V65/III, major (#2, #4, 6, 7) 0.250000 0.000011 0.0 % NaN NaN NaN NaN NaN

839 rows × 10 columns

chords_by_mode.apply_step("Counter")
count
mode corpus piece chord_and_mode scale_degrees
major beethoven_piano_sonatas 01-1 V7, major (5, 7, 2, 4) 11
I6, major (3, 5, 1) 8
I, major (1, 3, 5) 7
V, major (5, 7, 2) 5
V(64), major (5, 1, 3) 5
... ... ... ... ... ...
minor beethoven_piano_sonatas 32-2 #viio6(64), minor (3, 5, #7) 1
V(64), minor (5, 1, 3) 1
V(64)/vi, minor (3, 6, b1) 1
#viio43, minor (4, 6, #7, 2) 1
#viio2, minor (6, #7, 2, 4) 1

5018 rows × 1 columns

chords_by_mode.format = "scale_degree"
chords_by_mode.get_default_analysis().make_ranking_table()
mode major minor
chord_and_mode scale_degrees duration_qb proportion proportion_% chord_and_mode scale_degrees duration_qb proportion proportion_%
rank
1 I, major (1, 3, 5) 4589.083333 0.193962 19.4 % i, minor (1, 3, 5) 2007.333333 0.167231 16.72 %
2 V7, major (5, 7, 2, 4) 2106.750000 0.089044 8.9 % V, minor (5, #7, 2) 1157.875000 0.096462 9.65 %
3 V, major (5, 7, 2) 2081.250000 0.087966 8.8 % V7, minor (5, #7, 2, 4) 970.750000 0.080873 8.09 %
4 I6, major (3, 5, 1) 1362.625000 0.057593 5.76 % i6, minor (3, 5, 1) 536.375000 0.044685 4.47 %
5 V(64), major (5, 1, 3) 744.333333 0.031460 3.15 % V(64), minor (5, 1, 3) 516.000000 0.042988 4.3 %
... ... ... ... ... ... ... ... ... ... ...
835 vi7(11), major (6, 1, 3, 5) 0.250000 0.000011 0.0 % NaN NaN NaN NaN NaN
836 ii43(4), major (6, 1, 2, 5) 0.250000 0.000011 0.0 % NaN NaN NaN NaN NaN
837 vi7(4), major (6, 2, 3, 5) 0.250000 0.000011 0.0 % NaN NaN NaN NaN NaN
838 V7(64), major (5, 1, 3, 4) 0.250000 0.000011 0.0 % NaN NaN NaN NaN NaN
839 V65/III, major (#2, #4, 6, 7) 0.250000 0.000011 0.0 % NaN NaN NaN NaN NaN

839 rows × 10 columns

unigram_proportions.plot_grouped()