Datasets:
File size: 4,202 Bytes
9bd1d61 a272f55 9bd1d61 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 | # /// script
# requires-python = ">=3.10"
# dependencies = ["numpy>=1.26", "pandas>=2", "scikit-learn>=1.4"]
# ///
"""Simple scikit-learn genre classification baselines on FreeTZAN.
For every feature set written by scripts/extract_features.py and every classifier below:
features are standardized (fit on train), each hyperparameter setting is trained on train,
the setting with the best valid accuracy is kept (the first one listed on ties), and its
test accuracy is reported. Nothing is trained on valid or test. Everything is deterministic.
Output: resources/baselines.csv (one row per feature set x classifier) and a Markdown table
of test accuracies on stdout.
Usage: uv run scripts/run_baselines.py [--features-dir ~/.cache/freetzan/features]
"""
import argparse
from pathlib import Path
import numpy as np
import pandas as pd
from sklearn.base import clone
from sklearn.ensemble import RandomForestClassifier
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import ParameterGrid
from sklearn.neighbors import KNeighborsClassifier
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.svm import SVC
REPO = Path(__file__).resolve().parent.parent
FEATURES = {
'mfcc': 'MFCC mean/std',
'fma': 'FMA librosa features',
'clap': 'LAION-CLAP (music)',
'mert': 'MERT-v1-95M',
}
CLASSIFIERS = { # name -> (estimator, hyperparameter grid searched on valid)
'kNN': (KNeighborsClassifier(), {'n_neighbors': [1, 5, 10, 20]}),
'LogReg': (LogisticRegression(max_iter=10000), {'C': [1e-3, 1e-2, 1e-1, 1, 10]}),
'SVM (RBF)': (SVC(), {'C': [0.1, 1, 10, 100]}),
'Random forest': (RandomForestClassifier(n_estimators=500, random_state=0, n_jobs=-1), {}),
}
def main():
ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
ap.add_argument('--features-dir', default=str(Path.home() / '.cache' / 'freetzan' / 'features'))
ap.add_argument('--out', default=str(REPO / 'resources' / 'baselines.csv'))
args = ap.parse_args()
md = pd.read_csv(REPO / 'metadata.csv')
y = md.genre.to_numpy()
idx = {s: (md.split == s).to_numpy() for s in ('train', 'valid', 'test')}
rows = []
for feat, feat_name in FEATURES.items():
f = Path(args.features_dir) / f'{feat}.npz'
if not f.exists():
print(f'{feat}: {f} not found, skipping')
continue
z = np.load(f)
assert (z['track_id'] == md.track_id.to_numpy()).all(), f'{f} rows do not match metadata.csv'
X = z['X']
for clf_name, (clf, grid) in CLASSIFIERS.items():
best = None
for params in ParameterGrid(grid):
model = make_pipeline(StandardScaler(), clone(clf).set_params(**params))
model.fit(X[idx['train']], y[idx['train']])
valid_acc = model.score(X[idx['valid']], y[idx['valid']])
if best is None or valid_acc > best[0]:
best = (valid_acc, params, model)
valid_acc, params, model = best
test_acc = model.score(X[idx['test']], y[idx['test']])
rows.append(dict(features=feat, dim=X.shape[1], classifier=clf_name, params=str(params),
train_acc=round(model.score(X[idx['train']], y[idx['train']]), 3),
valid_acc=round(valid_acc, 3), test_acc=round(test_acc, 3)))
print(f'{feat:5s} {clf_name:14s} {str(params):22s} valid {valid_acc:.2f} test {test_acc:.2f}', flush=True)
res = pd.DataFrame(rows)
res.to_csv(args.out, index=False)
table = res.pivot(index='features', columns='classifier', values='test_acc')
table = table.loc[[f for f in FEATURES if f in table.index], list(CLASSIFIERS)]
dims = res.groupby('features').dim.first()
print(f'\nwrote {args.out}\n')
print('| Features | Dim | ' + ' | '.join(table.columns) + ' |')
print('|---|---:|' + '---:|' * len(table.columns))
for feat, r in table.iterrows():
print(f'| {FEATURES[feat]} | {dims[feat]} | ' + ' | '.join(f'{100 * v:.0f}%' for v in r) + ' |')
if __name__ == '__main__':
main()
|