Create midi_dataset.py
Browse files- midi_dataset.py +50 -0
midi_dataset.py
ADDED
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import pandas as pd
|
3 |
+
from datasets import DatasetBuilder, GeneratorBasedBuilder, Split, Value, Features, ClassLabel
|
4 |
+
|
5 |
+
class MidiDataset(GeneratorBasedBuilder):
|
6 |
+
"""Hugging Face Dataset for MIDI files and their labels"""
|
7 |
+
|
8 |
+
def _info(self):
|
9 |
+
return DatasetBuilder.info(
|
10 |
+
features=Features({
|
11 |
+
"file_path": Value("string"), # MIDI file path
|
12 |
+
"label": ClassLabel(names=["1", "2", "3", "4"]), # 4Q labels as class labels
|
13 |
+
"annotator": Value("string") # Annotator information
|
14 |
+
})
|
15 |
+
)
|
16 |
+
|
17 |
+
def _split_generators(self, dl_manager):
|
18 |
+
"""Split the dataset. Assumes all files are local."""
|
19 |
+
# Set paths to midis/ and label.csv
|
20 |
+
midi_path = "midis"
|
21 |
+
label_path = "label.csv"
|
22 |
+
|
23 |
+
return [
|
24 |
+
Split(
|
25 |
+
name=Split.TRAIN,
|
26 |
+
gen_kwargs={
|
27 |
+
"midi_dir": midi_path,
|
28 |
+
"label_file": label_path
|
29 |
+
}
|
30 |
+
)
|
31 |
+
]
|
32 |
+
|
33 |
+
def _generate_examples(self, midi_dir, label_file):
|
34 |
+
"""Yield examples from MIDI files and the label.csv"""
|
35 |
+
# Read the label.csv into a pandas DataFrame
|
36 |
+
df = pd.read_csv(label_file)
|
37 |
+
|
38 |
+
for index, row in df.iterrows():
|
39 |
+
midi_file = os.path.join(midi_dir, f"{row['ID']}.mid")
|
40 |
+
if os.path.exists(midi_file):
|
41 |
+
yield index, {
|
42 |
+
"file_path": midi_file,
|
43 |
+
"label": str(row["4Q"]), # Convert label to string for ClassLabel compatibility
|
44 |
+
"annotator": row["annotator"]
|
45 |
+
}
|
46 |
+
|
47 |
+
# Usage Example:
|
48 |
+
# You can now load the dataset using this script as follows:
|
49 |
+
# from datasets import load_dataset
|
50 |
+
# dataset = load_dataset("path/to/your/script", split="train")
|