Update wmms.py
Browse files
wmms.py
CHANGED
@@ -4,11 +4,13 @@
|
|
4 |
|
5 |
|
6 |
import os
|
|
|
7 |
import textwrap
|
8 |
import datasets
|
9 |
import itertools
|
10 |
import typing as tp
|
11 |
from pathlib import Path
|
|
|
12 |
from sklearn.model_selection import train_test_split
|
13 |
|
14 |
SAMPLE_RATE = 16_000
|
@@ -57,18 +59,40 @@ class WMMS(datasets.GeneratorBasedBuilder):
|
|
57 |
"""Returns SplitGenerators."""
|
58 |
archive_path = dl_manager.extract(_COMPRESSED_FILENAME)
|
59 |
extensions = ['.wav']
|
60 |
-
_,
|
|
|
61 |
|
62 |
-
train_walker, test_walker = train_test_split(
|
63 |
-
|
64 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
|
66 |
return [
|
67 |
datasets.SplitGenerator(
|
68 |
-
name=datasets.Split.TRAIN, gen_kwargs={"audio_paths":
|
69 |
),
|
70 |
datasets.SplitGenerator(
|
71 |
-
name=datasets.Split.TEST, gen_kwargs={"audio_paths":
|
72 |
),
|
73 |
]
|
74 |
|
|
|
4 |
|
5 |
|
6 |
import os
|
7 |
+
import random
|
8 |
import textwrap
|
9 |
import datasets
|
10 |
import itertools
|
11 |
import typing as tp
|
12 |
from pathlib import Path
|
13 |
+
from collections import defaultdict
|
14 |
from sklearn.model_selection import train_test_split
|
15 |
|
16 |
SAMPLE_RATE = 16_000
|
|
|
59 |
"""Returns SplitGenerators."""
|
60 |
archive_path = dl_manager.extract(_COMPRESSED_FILENAME)
|
61 |
extensions = ['.wav']
|
62 |
+
_, filepaths = fast_scandir(archive_path, extensions, recursive=True)
|
63 |
+
labels = [default_find_classes(f) for f in filepaths]
|
64 |
|
65 |
+
# train_walker, test_walker = train_test_split(
|
66 |
+
# _walker, test_size=0.2, random_state=914, stratify=[default_find_classes(f) for f in _walker]
|
67 |
+
# )
|
68 |
+
|
69 |
+
# Step 1: Organize samples by class
|
70 |
+
class_to_files = defaultdict(list)
|
71 |
+
for filepath, label in zip(filepaths, labels):
|
72 |
+
class_to_files[label].append(filepath)
|
73 |
+
|
74 |
+
# Step 2: Select exactly 2 samples per class for the test set
|
75 |
+
test_files, test_labels = [], []
|
76 |
+
train_files, train_labels = [], []
|
77 |
+
|
78 |
+
for label, files in class_to_files.items():
|
79 |
+
if len(files) < 2:
|
80 |
+
raise ValueError(f"Not enough samples for class {label}") # Ensure each class has at least 2 samples
|
81 |
+
|
82 |
+
random.Random(914).shuffle(files) # Shuffle to ensure randomness
|
83 |
+
|
84 |
+
test_files.extend(files[:2]) # Pick first 2 for test
|
85 |
+
test_labels.extend([label] * 2)
|
86 |
+
|
87 |
+
train_files.extend(files[2:]) # Remaining go to train
|
88 |
+
train_labels.extend([label] * (len(files) - 2))
|
89 |
|
90 |
return [
|
91 |
datasets.SplitGenerator(
|
92 |
+
name=datasets.Split.TRAIN, gen_kwargs={"audio_paths": train_files, "split": "train"}
|
93 |
),
|
94 |
datasets.SplitGenerator(
|
95 |
+
name=datasets.Split.TEST, gen_kwargs={"audio_paths": test_files, "split": "test"}
|
96 |
),
|
97 |
]
|
98 |
|