def stratified_fold(data_dir, annotation_file, n_folds=10, savefile=''):
modes = fileOperations.getModeNames(data_dir)
[filepaths, basefolders, filenames] = fileOperations.getFileNamesInDir(data_dir, extension='.pitch')
with open(annotation_file, 'r') as a:
annotations = json.load(a)
filemodes = [os.path.basename(b) for b in basefolders]
mbids = [os.path.splitext(f)[0] for f in filenames]
tonics = []
for m in mbids:
for a in annotations:
if a['mbid'] == m:
tonics.append(a['tonic'])
# get the stratified folds
mode_idx = [modes.index(m) for m in filemodes]
skf = cross_validation.StratifiedKFold(mode_idx, n_folds=n_folds, shuffle=True)
folds = dict()
for ff, fold in enumerate(skf):
folds['fold' + str(ff)] = {'train': [], 'test': []}
for tr_idx in fold[0]:
folds['fold' + str(ff)]['train'].append({'file': filepaths[tr_idx], 'mode': filemodes[tr_idx],
'tonic': tonics[tr_idx], 'mbid': mbids[tr_idx]})
for te_idx in fold[1]:
folds['fold' + str(ff)]['test'].append({'file': filepaths[te_idx], 'mode': filemodes[te_idx],
'tonic': tonics[te_idx], 'mbid': mbids[te_idx]})
# save the folds to a file if specified
if savefile:
with open(savefile, 'w') as f:
json.dump(folds, f, indent=2)
return folds
# coding: utf-8 import sys sys.path.insert(0, '../ModeTonicEstimation/') import json import os from extras import fileOperations as fo import numpy as np from ModeTonicEstimation import Chordia # I/O base_dir = '../../experiments/raga/' data_dir = os.path.join(base_dir,'data') experiments_dir = os.path.join(base_dir, 'experiments') modes = fo.getModeNames(data_dir) che = Chordia.Chordia(step_size=10, smooth_factor=15, chunk_size=0, threshold=0.5, overlap=0, frame_rate=128.0/44100) # indexing n_exp = 20 n_folds = 14 n_modes = len(modes) input_num = int(sys.argv[1])-1 [ex_idx, fo_idx, mode_idx] = np.unravel_index(input_num, [n_exp, n_folds, n_modes]) cur_mode = modes[mode_idx]