/
load_song_data.py
244 lines (197 loc) · 8.01 KB
/
load_song_data.py
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
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
import sys
import os
import json
import numpy as np
import matplotlib.pyplot as plt
from collections import Counter
import random
lib_path = os.path.abspath('./lib')
sys.path.append(lib_path)
import hdf5_getters
from utils import memoized
DATAPATH_ROOT = "./MSD-SHS/"
# Load the song data as 3 different dictionaries.
# track path gives a (key, value) as (<work>, <path>)
# track_info gives a (key, value) as (<work>, dictionary) where the keys to the dictionary
# are TID, AID, work, and clique_name
class Track:
def __init__(self, track_id, path, clique, DATAPATH='.'):
self.id = track_id
self.path = os.path.join(DATAPATH, path)
self.clique = clique
try:
self.h5 = hdf5_getters.open_h5_file_read(self.path)
self.h5.close()
self.h5 = None
except Exception as e:
self.h5 = -1
def close(self):
if self.h5 == None: return # do nothing
self.h5.close()
self.h5 = None
def open(self):
self.h5 = hdf5_getters.open_h5_file_read(self.path)
@memoized
def get_duration(self):
if self.h5 == None: self.open()
return hdf5_getters.get_duration(self.h5)
@memoized
def get_loudness(self):
if self.h5 == None: self.open()
return hdf5_getters.get_loudness(self.h5)
@memoized
def get_mode(self):
if self.h5 == None: self.open()
return hdf5_getters.get_mode(self.h5)
@memoized
def get_mode_confidence(self):
if self.h5 == None: self.open()
return hdf5_getters.get_mode_confidence(self.h5)
@memoized
def get_segments_loudness_max(self):
if self.h5 == None: self.open()
return hdf5_getters.get_segments_loudness_max(self.h5)
@memoized
def get_segments_loudness_max_time(self):
if self.h5 == None: self.open()
return hdf5_getters.get_segments_loudness_max_time(self.h5)
@memoized
def get_segments_loudness_start(self):
if self.h5 == None: self.open()
return hdf5_getters.get_segments_loudness_start(self.h5)
@memoized
def get_segments_pitches(self):
if self.h5 == None: self.open()
return hdf5_getters.get_segments_pitches(self.h5)
@memoized
def get_segments_timbre(self):
if self.h5 == None: self.open()
try:
timbre_by_segment = hdf5_getters.get_segments_timbre(self.h5)
except Exception as e:
print repr(e)
timbre_by_segment = None
return timbre_by_segment
@memoized
def get_tatums_start(self):
if self.h5 == None: self.open()
return hdf5_getters.get_tatums_start(self.h5)
@memoized
def get_tatums_confidence(self):
if self.h5 == None: self.open()
return hdf5_getters.get_tatums_confidence(self.h5)
@memoized
def get_tempo(self):
if self.h5 == None: self.open()
return hdf5_getters.get_tempo(self.h5)
@memoized
def get_time_signature(self):
if self.h5 == None: self.open()
return hdf5_getters.get_time_signature(self.h5)
@memoized
def get_time_signature_confidence(self):
if self.h5 == None: self.open()
return hdf5_getters.get_time_signature_confidence(self.h5)
@memoized
def get_key(self):
if self.h5 == None: self.open()
return hdf5_getters.get_key(self.h5)
@memoized
def get_danceability(self):
if self.h5 == None: self.open()
return hdf5_getters.get_danceability(self.h5)
@memoized
def get_energy(self):
if self.h5 == None: self.open()
return hdf5_getters.get_energy(self.h5)
@memoized
def get_segments_start(self):
if self.h5 == None: self.open()
return hdf5_getters.get_segments_start(self.h5)
@memoized
def get_beats_start(self):
if self.h5 == None: self.open()
return hdf5_getters.get_beats_start(self.h5)
@memoized
def get_tatums_start(self):
if self.h5 == None: self.open()
return hdf5_getters.get_tatums_start(self.h5)
class Track_dataset:
def __init__(self):
def get_dic(path):
json_data = open(path)
return json.load(json_data)
self.track_paths_full = get_dic(DATAPATH_ROOT + 'shs_dataset_full/shs_dataset_full.trackpaths.json')
self.track_info_full = get_dic(DATAPATH_ROOT + 'shs_dataset_full/shs_dataset_full.tracks.json')
self.track_paths_train = dict()
self.track_info_train = dict()
self.track_paths_test = dict()
self.track_info_test = dict()
self.track_paths_loc_train = get_dic(DATAPATH_ROOT + 'shs_dataset_train/shs_dataset_train.trackpaths.json')
self.track_paths_loc_test = get_dic(DATAPATH_ROOT + 'shs_dataset_test/shs_dataset_test.trackpaths.json')
def prune(self, ncliques=300, rseed=None, test_fraction=0.33):
"""Prune the dataset down to a smaller number of tracks."""
# get the count for cliques, and sort by size of the clique
clique_counter = Counter()
for k, v in self.track_info_full.iteritems():
clique_counter[v['clique_name']] += 1
sorted_clique_counter = dict(sorted(clique_counter.items(), key= lambda item: item[1], reverse=True)[:ncliques])
#print sorted_clique_counter
#trim down self.track_info_full
# get a dictionary of {clique_name : list of track_ids}
tmp_dic = dict()
clique_dic = dict()
for track_id, values in self.track_info_full.iteritems():
if values['clique_name'] in sorted_clique_counter:
tmp_dic[track_id] = values
if values['clique_name'] in clique_dic:
clique_dic[values['clique_name']].append(track_id)
else:
clique_dic[values['clique_name']] = [track_id]
self.track_info_full = tmp_dic
# split up training and testing sets. ~1/3 for testing (round down), and 2/3 for training (round up)
rng = random.Random()
rng.seed(rseed)
for clique, tracks in clique_dic.iteritems():
n_test = int(len(tracks)*test_fraction)
# random.shuffle(tracks)
rng.shuffle(tracks)
for i in xrange(len(tracks)):
if i < n_test:
self.track_info_test[tracks[i]] = self.track_info_full[tracks[i]]
else:
self.track_info_train[tracks[i]] = self.track_info_full[tracks[i]]
# populate self.track_paths
self.track_paths_train = {k:v for k,v in self.track_paths_full.items() if k in self.track_info_train}
self.track_paths_test = {k:v for k,v in self.track_paths_full.items() if k in self.track_info_test}
return clique_counter, sorted_clique_counter
def get_track(self, track_id):
def path_loc(track_id):
if track_id in self.track_paths_loc_train:
return 'train'
elif track_id in self.track_paths_loc_test:
return 'test'
else:
# should never go to this line
print "Error: not in train or test!"
return None
if track_id in self.track_info_train:
clique_name = self.track_info_train[track_id]['clique_name']
track_path = self.track_paths_train[track_id]
else:
clique_name = self.track_info_test[track_id]['clique_name']
track_path = self.track_paths_test[track_id]
return Track(track_id, track_path, clique_name, DATAPATH=os.path.join(DATAPATH_ROOT, path_loc(track_id)))
def get_tracks_train(self):
return self.get_tracks(self.track_paths_train)
def get_tracks_test(self):
return self.get_tracks(self.track_paths_test)
def get_tracks(self, paths):
tracks = []
for track_id in paths:
new_track = self.get_track(track_id)
if new_track.h5 == -1:
print "Error: unable to open track %s" % track_id
continue
tracks.append(new_track)
return tracks