-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
436 lines (307 loc) · 11 KB
/
utils.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
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
"""
Utility functions for handling tokens, intensities, etc.
TensorFlow and Midi-related libraries not needed here.
"""
import numpy as np
import os
import random
def split_token_to_arrays(tokens, split_number):
"""
Given a token list, `tokens`, turns the list into a list of lists of tokens using `split_number` as the token to split on.
Returns the tokens as a list of lists of tokens.
"""
output = []
output.append([])
split_number = str(split_number)
for token in tokens:
if token != split_number:
output[-1].append(token)
else:
output.append([])
return output
def join_token_array(tokens, join_number):
"""
Opposite of `split_token_to_arrays`. Joins a token list into a list with splits replaced with `join_number`.
Returns the joined list.
"""
output = []
for token in tokens:
output += token
output += [join_number]
return output
def remove_duplicates(tokens):
"""
Removes duplicates from a list of lists of tokens.
Returns the token array without duplicates per token list.
"""
output = []
for array in tokens:
new_array = []
for value in array:
if not (value in new_array):
new_array.append(value)
output.append(new_array)
return output
def get_tokens_between_beats(start_beat, end_beat, tokens):
"""
Given a list of lists of tokens (`tokens`) and a start beat and end beat,
calculates the tokens between the two beats.
Returns both the tokens between the two beat, as well as how many tokens are between the beats.
"""
indexes_to_take = end_beat - start_beat
this_tokens = []
this_tokens_total = 0
# print(start_beat + indexes_to_take)
# print(len(tokens))
if start_beat + indexes_to_take > len(tokens):
return [], 0
for index in range(indexes_to_take):
# print(len(tokens))
this_tokens.append(tokens[start_beat + index])
this_tokens_total += len(tokens[start_beat + index])
# print(tokens[this_beat + index])
return this_tokens, this_tokens_total
def combine_tokens(tokens):
"""
Combines an list of token lists into a single token list.
Returns the combined list.
"""
max_length = -1
for item in tokens:
if len(item) > max_length:
max_length = len(item)
combined = []
for index in range(max_length):
combined.append([])
for row in tokens:
if index >= len(row):
continue
combined[-1] += (row[index])
output = combined
return output
def tokens_are_empty(tokens):
"""
Check if a token array/list is actually empty.
i.e., `[[], [], []]` is technically not empty, but for our purposes, it is.
Returns whether it is empty.
"""
if len(tokens) == 0:
return True
for token in tokens:
if token != 0:
return False
return True
def get_random_cut(inputs, length):
"""
Gets a random cut, or 'window', from `inputs` of length `length`.
Returns the cut as a list of length `length`.
"""
selected = random.choice(inputs)
start = int(np.random.uniform(0, len(selected) - length))
output = []
for item in range(length):
output.append(selected[start + item])
return output
def pad_input(input, buffer_size):
"""
Pads an input formatted as a list of token lists, `input`, to the given size, `buffer_size`.
Padded with empty lists, `[]`, at the front.
Returns the padded input.
"""
if len(input) < (buffer_size - 1):
print("Padding: " + str(input))
input = [[]] * ((buffer_size - 1) - len(input)) + input
print("Padded to: " + str(input))
return input
def get_pitches_from_output(prediction, dictionary, cutoff=0.5, num_to_force=0):
"""
Get pitches from the model's output (`prediction`), converting to pitches using `dictionary`.
`cutoff` determines how activate a unit has to be (between `0` and `1`) to be considered a valid output.
Optionally can force `num_to_force` outputs to be valid. The most active are forced first.
Returns the output pitches as a list.
"""
output = []
prediction = prediction[0]
# handle cutoff
for idx in range(len(prediction)):
confidence = prediction[idx]
if confidence >= cutoff:
pitch = next(pitch for pitch, value in dictionary.items() if value == idx)
output.append(str(pitch))
# handle forced
forced = 0
while (forced != num_to_force) and (len(prediction) >= 0):
max_value = np.argmax(prediction)
prediction = np.delete(prediction, max_value)
pitch = next(pitch for pitch, value in dictionary.items() if value == max_value)
forced += 1
if str(pitch) in output:
continue
output.append(str(pitch))
return output
def generate_random_input(buffer_size, token_count):
"""
Generates a random input.
(This will likely not yield very good results...)
"""
random_inputs = []
for x in range(buffer_size - 1):
generated = np.random.randint(1, token_count, np.random.randint(1, 5, 1)).tolist()
generated = list(map(str, generated))
random_inputs.append(generated)
return random_inputs
def generate_basic_drums(buffer_size, token_count):
"""
Generates a basic drum pattern as an input.
1. Bass drum
2. Hi-hat
3. Snare
4. Hi-hat
5. Repeat...
"""
programmed_input = []
drums_length = int((buffer_size - 1) // 4) * 4
for x in range(drums_length):
programmed_input.append(["i0"])
# force drums on beats 1, 3, and snare on 2, 4, hh on off beats
for x in range(len(programmed_input)):
if (x % 4) == 0: # kick
programmed_input[x].append("35")
if ((x + 2) % 4) == 0: # snare
programmed_input[x].append("38")
programmed_input[x].append("42") # hh
programmed_input = pad_input(programmed_input, buffer_size)
return programmed_input
def generate_blank_input(buffer_size, token_count):
"""
Generates a completely blank input.
"""
return [[]]
def generate_count_in(buffer_size, token_count):
"""
Generates a count in of four hi-hats.
"""
programmed_input = [["42"], [], ["42"], [], ["42"], [], ["42"], []]
for i in programmed_input:
i.append("i-2")
return programmed_input
def categorical_encoder(tokens, num_classes, dictionary):
"""
A categorical encoder function to turn `tokens` into categorically encoded classes of
total `num_classes` using `dictionary` to look up class encodings.
Returns the tokens categorically encoded as a `numpy` array.
"""
array = np.zeros(shape=(len(tokens), num_classes))
for token_array_idx in range(len(tokens)):
token_array = tokens[token_array_idx]
for token_idx in range(len(token_array)):
token = token_array[token_idx]
if "i" in token:
array[token_array_idx][dictionary[-999]] = float(token.replace("i", ""))
else:
token = int(token)
if token in dictionary:
array[token_array_idx][dictionary[token]] = 1
return array
def input_to_categories(input, token_count, buffer_size, dictionary):
"""
Converts an input (list of tokens) into a categories for inputting into the neural network.
Returns the input in category encoding as a valid network input.
"""
input = pad_input(input, buffer_size)
new_input = categorical_encoder(input, token_count, dictionary)
new_input = new_input.reshape(-1, buffer_size - 1, token_count)
return new_input
def intensities_to_float(tokens):
"""
Converts intensity tokens as a list of strings into a list of floats.
Ignores non-intensity tokens.
Returns the list of floats.
"""
intensities = []
for beat in tokens:
for token in beat:
if "i" in token:
intensities.append(float(token.replace("i", "")))
return intensities
def extract_intensities(tokens):
"""
Extracts only the intensities from a given token list.
Returns just the intensities as a list seperated by timestep.
"""
intensities = []
for beat in tokens:
for token in beat:
if "i" in token:
intensities.append([token])
return intensities
def remove_intensity_outliers(intensity_array, mean, std):
"""
Removes intensity outliers by replacing them with the closest non-outlier intensity.
Outliers are defined by being either bigger or smaller than `mean +/- (2 * std)`.
Returns a list of intensities with outliers removed.
"""
new_array = []
for intensity in intensity_array:
if intensity < mean + (2 * std):
if intensity > mean - (2 * std):
new_array.append(intensity)
continue
new_array.append(-999)
newer_array = []
for index in range(len(new_array)):
intensity = new_array[index]
if intensity != -999:
newer_array.append(intensity)
continue
prev_actual = -999
search_index = index
while (prev_actual == -999 and search_index >= 0):
this_intensity = new_array[search_index]
if this_intensity != -999:
prev_actual = this_intensity
break
search_index -= 1
next_actual = -999
search_index = index
while (next_actual == -999 and search_index < len(new_array)):
this_intensity = new_array[search_index]
if this_intensity != -999:
next_actual = this_intensity
break
search_index += 1
if prev_actual == -999:
prev_actual = next_actual
if next_actual == -999:
next_actual = prev_actual
average = np.mean([prev_actual, next_actual])
newer_array.append(average)
return newer_array
def normalize_intensities(intensity_tokens):
"""
Normalize intensities by removing outliers and then subtracting the mean and dividing by the standard deviation.
Returns normalized intensities.
"""
unpacked = []
for beat in intensity_tokens:
intensity = float(beat[0].replace("i", ""))
unpacked.append(intensity)
mean = np.mean(unpacked)
std = np.std(unpacked)
intensity_tokens = remove_intensity_outliers(unpacked, mean, std)
# recalc without extremes
mean = np.mean(intensity_tokens)
std = np.std(intensity_tokens)
new_array = []
for beat in intensity_tokens:
intensity = beat
intensity -= mean
intensity /= std
new_array.append(["i" + str(round(intensity, 1))])
return new_array
def clean_temp():
"""
Clean up temporary files.
"""
if os.path.exists("temp.mid"):
os.remove("temp.mid")