trojblue commited on
Commit
6fd7f2e
·
verified ·
1 Parent(s): df9cc8d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +76 -0
README.md CHANGED
@@ -161,3 +161,79 @@ This dataset is a structured extraction of the [Million Song Subset](http://mill
161
  For more details, visit the [Million Song Dataset website](http://millionsongdataset.com).
162
 
163
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
161
  For more details, visit the [Million Song Dataset website](http://millionsongdataset.com).
162
 
163
 
164
+ ## Appendix: Processing Code
165
+
166
+ The dataset was converted using the following snippet:
167
+
168
+ ```python
169
+ import os
170
+ import unibox as ub
171
+ import pandas as pd
172
+ import numpy as np
173
+ import h5py
174
+ from tqdm import tqdm
175
+ from concurrent.futures import ProcessPoolExecutor
176
+
177
+ # https://github.com/tbertinmahieux/MSongsDB/blob/0c276e289606d5bd6f3991f713e7e9b1d4384e44/PythonSrc/hdf5_getters.py
178
+ import hdf5_getters
179
+
180
+ # Define dataset path
181
+ dataset_path = "/lv0/yada/dataproc5/data/MillionSongSubset"
182
+
183
+ # Function to extract all available fields from an HDF5 file
184
+ def extract_song_data(file_path):
185
+ """Extracts all available fields from an HDF5 song file using hdf5_getters."""
186
+ song_data = {}
187
+
188
+ try:
189
+ with hdf5_getters.open_h5_file_read(file_path) as h5:
190
+ # Get all getter functions from hdf5_getters
191
+ getters = [func for func in dir(hdf5_getters) if func.startswith("get_")]
192
+
193
+ for getter in getters:
194
+ try:
195
+ # Dynamically call each getter function
196
+ value = getattr(hdf5_getters, getter)(h5)
197
+
198
+ # Optimize conversions
199
+ if isinstance(value, np.ndarray):
200
+ value = value.tolist()
201
+ elif isinstance(value, bytes):
202
+ value = value.decode()
203
+
204
+ # Store in dictionary with a cleaned-up key name
205
+ song_data[getter[4:]] = value
206
+
207
+ except Exception:
208
+ continue # Skip errors but don't slow down
209
+
210
+ except Exception as e:
211
+ print(f"Error processing {file_path}: {e}")
212
+
213
+ return song_data
214
+
215
+ # Function to process multiple files in parallel
216
+ def process_files_in_parallel(h5_files, num_workers=8):
217
+ """Processes multiple .h5 files in parallel."""
218
+ all_songs = []
219
+
220
+ with ProcessPoolExecutor(max_workers=num_workers) as executor:
221
+ for song_data in tqdm(executor.map(extract_song_data, h5_files), total=len(h5_files)):
222
+ if song_data:
223
+ all_songs.append(song_data)
224
+
225
+ return all_songs
226
+
227
+ # Find all .h5 files
228
+ h5_files = [os.path.join(root, file) for root, _, files in os.walk(dataset_path) for file in files if file.endswith(".h5")]
229
+
230
+ # Process files in parallel
231
+ all_songs = process_files_in_parallel(h5_files, num_workers=24)
232
+
233
+ # Convert to Pandas DataFrame
234
+ df = pd.DataFrame(all_songs)
235
+
236
+ ub.saves(df, "hf://trojblue/million-song-subset", private=False)
237
+ ```
238
+
239
+