| jupytext | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
|
import os
from collections import defaultdict, Counter
from git import Repo
import dimcat as dc
import ms3
import pandas as pd
import plotly.express as px
import plotly.graph_objects as go
from utils import STD_LAYOUT, CADENCE_COLORS, color_background, value_count_df, get_repo_name, resolve_dir
CORPUS_PATH = os.environ.get('CORPUS_PATH', "~/dcml_corpora")
CORPUS_PATH = resolve_dir(CORPUS_PATH)
repo = Repo(CORPUS_PATH)
notebook_repo = Repo('.', search_parent_directories=True)
print(f"Notebook repository '{get_repo_name(notebook_repo)}' @ {notebook_repo.commit().hexsha[:7]}")
print(f"Data repo '{get_repo_name(CORPUS_PATH)}' @ {repo.commit().hexsha[:7]}")
print(f"dimcat version {dc.__version__}")
print(f"ms3 version {ms3.__version__}")
dataset = dc.Dataset()
dataset.load(directory=CORPUS_PATH)
dataset.data
hascadence = dc.HasCadenceAnnotationsFilter().process_data(dataset)
print(f"Before: {len(dataset.indices[()])} pieces; after removing those without cadence labels: {len(hascadence.indices[()])}")
grouped_by_dataset = dc.CorpusGrouper().process_data(hascadence)
corpora = {group[0]: f"{len(ixs)} pieces" for group, ixs in grouped_by_dataset.indices.items()}
print(f"{len(corpora)} corpora with {sum(map(len, grouped_by_dataset.indices.values()))} pieces containing cadence annotations:")
corpora
all_labels = hascadence.get_facet('expanded')
print(f"{len(all_labels.index)} hand-annotated harmony labels:")
all_labels.iloc[:10, 14:].style.apply(color_background, subset="chord")
dataset_metadata = hascadence.data.metadata()
hascadence_metadata = dataset_metadata.loc[hascadence.indices[()]]
hascadence_metadata.index.rename('dataset', level=0, inplace=True)
hascadence_metadata.head()
mean_composition_years = hascadence_metadata.groupby(level=0).composed_end.mean().astype(int).sort_values()
chronological_order = mean_composition_years.index.to_list()
bar_data = pd.concat([mean_composition_years.rename('year'),
hascadence_metadata.groupby(level='dataset').size().rename('pieces')],
axis=1
).reset_index()
fig = px.bar(bar_data, x='year', y='pieces', color='dataset', title='Pieces contained in the dataset')
fig.update_traces(width=5)
- PAC: Perfect Authentic Cadence
- IAC: Imperfect Authentic Cadence
- HC: Half Cadence
- DC: Deceptive Cadence
- EC: Evaded Cadence
- PC: Plagal Cadence
print(f"{all_labels.cadence.notna().sum()} cadence labels.")
value_count_df(all_labels.cadence)
px.pie(all_labels[all_labels.cadence.notna()], names="cadence", color="cadence", color_discrete_map=CADENCE_COLORS)
cadence_count_per_dataset = all_labels.groupby("corpus").cadence.value_counts()
cadence_fraction_per_dataset = cadence_count_per_dataset / cadence_count_per_dataset.groupby(level=0).sum()
px.bar(cadence_fraction_per_dataset.rename('count').reset_index(), x='corpus', y='count', color='cadence',
color_discrete_map=CADENCE_COLORS, category_orders=dict(dataset=chronological_order))
fig = px.pie(cadence_count_per_dataset.rename('count').reset_index(), names='cadence', color='cadence', values='count',
facet_col='corpus', facet_col_wrap=4, height=2000, color_discrete_map=CADENCE_COLORS)
fig.for_each_annotation(lambda a: a.update(text=a.text.split("=")[-1]))
fig.update_layout(**STD_LAYOUT)
:tags: []
segmented = dc.PhraseSlicer().process_data(grouped_by_dataset)
phrases = segmented.get_slice_info()
phrase_segments = segmented.get_facet("expanded")
phrase_gpb = phrase_segments.groupby(level=[0,1,2])
local_keys_per_phrase = phrase_gpb.localkey.unique().map(tuple)
n_local_keys_per_phrase = local_keys_per_phrase.map(len)
phrases_with_keys = pd.concat([n_local_keys_per_phrase.rename('n_local_keys'),
local_keys_per_phrase.rename('local_keys'),
phrases], axis=1)
phrases_with_cadences = pd.concat([
phrase_gpb.cadence.nunique().rename('n_cadences'),
phrase_gpb.cadence.unique().rename('cadences').map(lambda l: tuple(e for e in l if not pd.isnull(e))),
phrases_with_keys
], axis=1)
value_count_df(phrases_with_cadences.n_cadences, counts="#phrases")
n_cad = phrases_with_cadences.groupby(level='corpus').n_cadences.value_counts().rename('counts').reset_index().sort_values('n_cadences')
n_cad.n_cadences = n_cad.n_cadences.astype(str)
fig = px.bar(n_cad, x='corpus', y='counts', color='n_cadences', height=800, barmode='group',
labels=dict(n_cadences="#cadences in a phrase"),
category_orders=dict(dataset=chronological_order)
)
fig.show()
value_count_df(phrases_with_cadences[phrases_with_cadences.n_cadences > 1].cadences)
df_rows = []
y_position = 0
for ix in phrases_with_cadences[phrases_with_cadences.n_cadences > 0].sort_values('duration_qb').index:
df = phrase_segments.loc[ix]
description = str(ix)
if df.cadence.notna().any():
interval = ix[2]
df_rows.append((y_position, interval.length, "end of phrase", description))
start_pos = interval.left
cadences = df.loc[df.cadence.notna(), ['quarterbeats', 'cadence']]
cadences.quarterbeats -= start_pos
for cadence_x, cadence_type in cadences.itertuples(index=False, name=None):
df_rows.append((y_position, cadence_x, cadence_type, description))
y_position += 1
#else:
# df_rows.append((y_position, pd.NA, pd.NA, description))
data = pd.DataFrame(df_rows, columns=["phrase_ix", "x", "marker", "description"])
fig = px.scatter(data[data.x.notna()], x='x', y="phrase_ix", color="marker", hover_name="description", height=3000,
labels=dict(marker='legend'), color_discrete_map=CADENCE_COLORS)
fig.update_traces(marker_size=5)
fig.update_yaxes(autorange="reversed")
fig.show()
phrase_segments = segmented.get_facet("expanded")
cadence_selector = phrase_segments.cadence.notna()
missing_chord_selector = phrase_segments.chord.isna()
cadence_with_missing_chord_selector = cadence_selector & missing_chord_selector
missing = phrase_segments[cadence_with_missing_chord_selector]
expanded = ms3.expand_dcml.expand_labels(phrase_segments[cadence_with_missing_chord_selector], propagate=False, chord_tones=True, skip_checks=True)
phrase_segments.loc[cadence_with_missing_chord_selector] = expanded
print(f"Ultima harmony missing for {(phrase_segments.cadence.notna() & phrase_segments.bass_note.isna()).sum()} cadence labels.")
def highlight(row, color="#ffffb3"):
if row.counts < 10:
return [None, None, None, None]
else:
return ["background-color: {color};"] * 4
cadence_counts = all_labels.cadence.value_counts()
ultima_root = phrase_segments.groupby(['localkey_is_minor', 'cadence']).numeral.value_counts().rename('counts').to_frame().reset_index()
ultima_root.localkey_is_minor = ultima_root.localkey_is_minor.map({False: 'in major', True: 'in minor'})
#ultima_root.style.apply(highlight, axis=1)
fig = px.pie(ultima_root, names='numeral', values='counts',
facet_row='cadence', facet_col='localkey_is_minor',
height=1500,
category_orders={'cadence': cadence_counts.index},
)
fig.for_each_annotation(lambda a: a.update(text=a.text.split("=")[-1]))
fig.update_traces(textposition='inside', textinfo='percent+label')
fig.update_layout(**STD_LAYOUT)
fig.show()
#phrase_segments.groupby(level=[0,1,2], group_keys=False).apply(lambda df: df if ((df.cadence == 'PAC') & (df.numeral == 'V')).any() else None)
ultima_bass = phrase_segments.groupby(['localkey_is_minor','cadence']).bass_note.value_counts().rename('counts').reset_index()
ultima_bass.bass_note = ms3.transform(ultima_bass, ms3.fifths2sd, dict(fifths='bass_note', minor='localkey_is_minor'))
ultima_bass.localkey_is_minor = ultima_bass.localkey_is_minor.map({False: 'in major', True: 'in minor'})
#ultima_bass.style.apply(highlight, axis=1)
fig = px.pie(ultima_bass, names='bass_note', values='counts',
facet_row='cadence', facet_col='localkey_is_minor',
height=1500,
category_orders={'cadence': cadence_counts.index},
)
fig.for_each_annotation(lambda a: a.update(text=a.text.split("=")[-1]))
fig.update_traces(textposition='inside', textinfo='percent+label')
fig.update_layout(**STD_LAYOUT)
fig.show()
+++
#pac_on_i = phrase_segments.groupby(level=[0,1,2], group_keys=False).apply(lambda df: df if ((df.cadence == 'PAC') & (df.numeral.isin(('I', 'i')))).any() else None)
#pac_on_i.cadence.value_counts()
#pac_on_i.droplevel(-1).index.nunique()
def get_progressions(selected='PAC', last_row={}, feature='chord', dataset=None, as_series=True):
"""Uses the nonlocal variable phrase_segments."""
last_row = {k: v if isinstance(v, tuple) else (v,) for k, v in last_row.items()}
progressions = []
for (corp, fname, *_), df in phrase_segments[phrase_segments[feature].notna()].groupby(level=[0,1,2]):
if dataset is not None and dataset not in corp:
continue
if (df.cadence == selected).fillna(False).any():
# remove chords after the last cadence label
df = df[df.cadence.fillna(method='bfill').notna()]
# group segments leading up to a cadence label
cadence_groups = df.cadence.notna().shift().fillna(False).cumsum()
for i, cadence in df.groupby(cadence_groups):
last_r = cadence.iloc[-1]
typ = last_r.cadence
if typ != selected:
continue
if any(last_r[feat] not in values for feat, values in last_row.items()):
continue
progressions.append(tuple(cadence[feature]))
if as_series:
return pd.Series(progressions)
return progressions
chord_progressions = get_progressions('PAC', dict(numeral=('I', 'i')), 'chord')
print(f"Progressions for {len(chord_progressions)} cadences:")
value_count_df(chord_progressions, "chord progressions")
numeral_progressions = get_progressions('PAC', dict(numeral=('I', 'i')), 'numeral')
value_count_df(numeral_progressions, "numeral progressions")
def remove_immediate_duplicates(l):
return tuple(a for a, b in zip(l, (None, ) + l) if a != b)
numeral_prog_no_dups = numeral_progressions.map(remove_immediate_duplicates)
value_count_df(numeral_prog_no_dups)
Scale degrees expressed w.r.t. major scale, regardless of actual key.
bass_progressions = get_progressions('PAC', dict(bass_note=0), 'bass_note')
bass_prog = bass_progressions.map(ms3.fifths2sd)
print(f"Progressions for {len(bass_progressions)} cadences:")
value_count_df(bass_prog, "bass progressions")
bass_prog_no_dups = bass_prog.map(remove_immediate_duplicates)
value_count_df(bass_prog_no_dups)
def make_sankey(data, labels, node_pos=None, margin={'l': 10, 'r': 10, 'b': 10, 't': 10}, pad=20, color='auto', **kwargs):
if color=='auto':
unique_labels = set(labels)
color_step = 100 / len(unique_labels)
unique_colors = {label: f'hsv({round(i*color_step)}%,100%,100%)' for i, label in enumerate(unique_labels)}
color = list(map(lambda l: unique_colors[l], labels))
fig = go.Figure(go.Sankey(
arrangement = 'snap',
node = dict(
pad = pad,
#thickness = 20,
#line = dict(color = "black", width = 0.5),
label = labels,
x = [node_pos[i][0] if i in node_pos else 0 for i in range(len(labels))] if node_pos is not None else None,
y = [node_pos[i][1] if i in node_pos else 0 for i in range(len(labels))] if node_pos is not None else None,
color = color,
),
link = dict(
source = data.source,
target = data.target,
value = data.value
),
),
)
fig.update_layout(margin=margin, **kwargs)
return fig
def progressions2graph_data(progressions, cut_at_stage=None):
stage_nodes = defaultdict(dict)
edge_weights = Counter()
node_counter = 0
for progression in progressions:
previous_node = None
for stage, current in enumerate(reversed(progression)):
if cut_at_stage and stage > cut_at_stage:
break
if current in stage_nodes[stage]:
current_node = stage_nodes[stage][current]
else:
stage_nodes[stage][current] = node_counter
current_node = node_counter
node_counter += 1
if previous_node is not None:
edge_weights.update([(current_node, previous_node)])
previous_node = current_node
return stage_nodes, edge_weights
def graph_data2sankey(stage_nodes, edge_weights):
data = pd.DataFrame([(u, v, w) for (u, v), w in edge_weights.items()], columns = ['source', 'target', 'value'])
node2label = {node: label for stage, nodes in stage_nodes.items() for label, node in nodes.items()}
labels = [node2label[i] for i in range(len(node2label))]
return make_sankey(data, labels)
def plot_progressions(progressions, cut_at_stage=None):
stage_nodes, edge_weights = progressions2graph_data(progressions, cut_at_stage=cut_at_stage)
return graph_data2sankey(stage_nodes, edge_weights)
plot_progressions(numeral_prog_no_dups, cut_at_stage=3)
chord_progressions_minor = get_progressions('PAC', dict(numeral='i', localkey_is_minor=True), 'root')
chord_progressions_minor
pac_major = get_progressions('PAC', dict(numeral='I', localkey_is_minor=False), 'chord')
plot_progressions(pac_major, cut_at_stage=4)
deceptive = get_progressions('DC', dict(localkey_is_minor=False), 'chord')
deceptive.value_counts()
plot_progressions(deceptive, cut_at_stage=4)
plot_progressions(bass_prog_no_dups, cut_at_stage=7)
def remove_sd_accidentals(t):
return tuple(map(lambda sd: sd[-1], t))
bass_prog_no_acc_no_dup = bass_prog.map(remove_sd_accidentals).map(remove_immediate_duplicates)
plot_progressions(bass_prog_no_acc_no_dup, cut_at_stage=7)
half = get_progressions('HC', dict(numeral='V'), 'bass_note').map(ms3.fifths2sd)
print(f"Progressions for {len(half)} cadences:")
plot_progressions(half.map(remove_immediate_duplicates), cut_at_stage=5)