def get_example(self, i):
if self.mix: # Training phase of BC learning
# Select two training examples
while True:
sound1, label1 = self.base[random.randint(
0,
len(self.base) - 1)]
sound2, label2 = self.base[random.randint(
0,
len(self.base) - 1)]
if label1 != label2:
break
sound1 = self.preprocess(sound1)
sound2 = self.preprocess(sound2)
# Mix two examples
r = np.array(random.random())
sound = U.mix(sound1, sound2, r, self.opt.fs).astype(np.float32)
eye = np.eye(self.opt.nClasses)
label = (eye[label1] * r + eye[label2] * (1 - r)).astype(
np.float32)
else: # Training phase of standard learning or testing phase
sound, label = self.base[i]
sound = self.preprocess(sound).astype(np.float32)
label = np.array(label, dtype=np.int32)
if self.train and self.opt.strongAugment:
sound = U.random_gain(6)(sound).astype(np.float32)
return sound, label
def audio_mix():
if globals.StartFlag and \
len(audio.output_queue) < MAX_AUDIO_OUTPUT_QUEUE_LENGTH:
end = globals.OutputSequenceNumber + BLOCK_SIZE / FRAMES_PER_PACKET
data = []
volumes = []
for player in globals.Players:
if player.id == globals.myPlayer.id:
continue
missing_data = False
for i in xrange(globals.OutputSequenceNumber, end):
if i not in player.audio_packets:
missing_data = True
if missing_data:
continue
d = ''
for i in xrange(globals.OutputSequenceNumber, end):
d += player.audio_packets[i]
data.append(d)
volumes.append(player.volume)
print 'eating: %d through %d' % (globals.OutputSequenceNumber, end)
globals.OutputSequenceNumber = end
audio.output_queue.append(mix(data, volumes))
def get_train_example(iterator_next):
'''Get a training or testing sample
'''
if mix: # Training phase of BC learning
# sound1, label1, sound2, label2 = U.get_different_sounds(iterator_next)
sound1, label1 = iterator_next
sound2, label2 = iterator_next
ratio = np.array(np.random.random())
sound = U.mix(sound1, sound2, ratio, opt.fs)
eye = tf.eye(opt.nClasses)
label = tf.gather(eye, label1) * ratio
label += tf.gather(eye, label2) * (1 - ratio)
# elif not train and opt.longAudio > 0: # Mix two audio on long frame (for testing)
# sound1, label1, sound2, label2 = U.get_different_sounds(iterator)
# raise NotImplementedError()
# Training phase of standard learning or testing phase
# sound, label = iterator_next
if opt.noiseAugment:
sound_len = tf.shape(sound)[0]
sound = sound + 0.01 * next_noise(is_train, sound_len)
if train and opt.strongAugment:
sound, label = U.random_gain(6)(sound, label)
return sound, label
def run():
global seq_num
end = seq_num + BLOCK_SIZE / FRAMES_PER_PACKET
data = []
volumes = []
if end in players[0].audio_packets:
for player in players:
d = ''
for i in xrange(seq_num, end):
d += player.audio_packets[i]
data.append(d)
volumes.append(player.volume)
audio.output_queue.append((seq_num, end, mix(data, volumes)))
seq_num = end
def close():
dump_file = wave.open('test_out.wav', 'w')
dump_file.setparams((2, 2, 44100, 0, 'NONE', 'not compressed'))
while len(audio.input_queue):
seq, data = audio.input_queue.pop(0)
input_audio_packets[seq] = data
to_write = []
for seq in sorted(input_audio_packets.keys()):
volumes = [p.volume for p in players]
data = [p.audio_packets[seq] for p in players]
volumes.append(10.0)
data.append(input_audio_packets[seq])
to_write.append(mix(data, volumes))
dump_file.writeframes(''.join(to_write))
dump_file.close()
from theano import tensor as T
PURCENT = 5 # Purcentage of the set you want on the test set
NUM_FRAMES = 60
DATADIR = '/baie/corpus/emoMusic/train/'
# DATADIR = './train/'
do_regularize = False
y_, song_id, nb_of_songs = load_y(DATADIR)
X_ = load_X(DATADIR, song_id)
# Now let's mix everything so that we can take test_set and train_set independantly
# We need to separate PER SONG
X_train, y_train, X_test, y_test, song_id_tst = mix(X_, y_, PURCENT,
NUM_FRAMES, song_id,
nb_of_songs)
print X_train.shape, y_train.shape, X_test.shape, y_test.shape
# print X_train[0:3,0:3]
# standardize data
X_train, scaler = standardize(X_train)
X_test, _ = standardize(X_test, scaler)
X_train = X_train[:, [
10, 12, 13, 17, 19, 82, 83, 84, 85, 89, 90, 91, 103, 140, 142, 146, 148,
212, 214, 218, 220
]]
X_test = X_test[:, [
10, 12, 13, 17, 19, 82, 83, 84, 85, 89, 90, 91, 103, 140, 142, 146, 148,
212, 214, 218, 220
import theano
from theano import tensor as T
PURCENT = 5 # Purcentage of the set you want on the test set
NUM_FRAMES = 60
DATADIR = "/baie/corpus/emoMusic/train/"
# DATADIR = './train/'
do_regularize = False
y_, song_id, nb_of_songs = load_y(DATADIR)
X_ = load_X(DATADIR, song_id)
# Now let's mix everything so that we can take test_set and train_set independantly
# We need to separate PER SONG
X_train, y_train, X_test, y_test, song_id_tst = mix(X_, y_, PURCENT, NUM_FRAMES, song_id, nb_of_songs)
print X_train.shape, y_train.shape, X_test.shape, y_test.shape
# print X_train[0:3,0:3]
# print np.mean(X_train[:,0:3], axis=0), np.std(X_train[:,0:3], axis=0)
# print np.mean(X_test[:,0:3], axis=0), np.std(X_test[:,0:3], axis=0)
# with(open('train_dummy.txt', mode='w')) as infile:
# for i in range(X_train.shape[0]):
# s=''
# for feat in range(3):
# s = s + '%g '%X_train[i,feat]
# infile.write('%s\n'%s)
# standardize data
X_train, scaler = standardize(X_train)
X_test, _ = standardize(X_test, scaler)
def show_results(topology, results, size, verbose):
for result in results:
items = list()
border = np.concatenate([
np.zeros((2 * size, 2, 3)), 0.5 * np.ones((2 * size, 2 * 2, 3)),
np.zeros((2 * size, 2, 3))
],
axis=1)
name = 'input'
if 'name' in result:
name += ': ' + str(result['name'], 'utf-8')
if 'variant' in result:
name += '(' + str(result['variant'], 'utf-8') + ')'
# s = set([
# 'input: image_08046(0.75_none_0.0_0)',
# 'input: image_07810(0.75_none_0.0_1)',
# 'input: image_07119(0.75_none_0.0_3)',
# ])
# if name not in s:
# continue
inputs = normalize(cv2.cvtColor(result['inputs'], cv2.COLOR_RGB2BGR))
items.append(annotate(inputs, size, 'input', name))
if 'labels' in result:
labels = cv2.cvtColor(result['labels'], cv2.COLOR_GRAY2BGR)
items.append(border)
items.append(annotate(labels, size, 'label'))
predictions = cv2.cvtColor(result['predictions'].astype(np.float32),
cv2.COLOR_GRAY2BGR)
items.append(border)
items.append(annotate(predictions, size, 'prediction', topology))
if 'mosaic' in result:
r = result['mosaic'][:, :, 0]
g = result['mosaic'][:, :, 1]
b = result['mosaic'][:, :, 2]
tp = np.sum(
np.bitwise_and(np.bitwise_and(r == 0, g == 255), b == 0))
fp = np.sum(
np.bitwise_and(np.bitwise_and(r == 255, g == 0), b == 0))
tn = np.sum(np.bitwise_and(np.bitwise_and(r == 0, g == 0), b == 0))
fn = np.sum(
np.bitwise_and(np.bitwise_and(r == 255, g == 0), b == 255))
if tp + fp > 0:
sp = tp / (tp + fp)
else:
sp = 1
if tp + fn > 0:
sr = tp / (tp + fn)
else:
sr = 1
if sp + sr > 0:
f1 = 2 * (sp * sr) / (sp + sr)
else:
f1 = 0
items.append(border)
items.append(
annotate(
cv2.cvtColor(tofloats(result['mosaic']),
cv2.COLOR_RGB2BGR), size,
'iou (P:%.00f%%, R:%.00f%%, F1:%.02f)' %
(100 * sp, 100 * sr, f1), topology))
if 'labels' in result:
items.append(border)
items.append(annotate(mix(inputs, labels), size, 'target'))
items.append(border)
items.append(
annotate(mix(inputs, predictions), size, 'result', topology))
cv2.imshow('cv2: result', tobytes(np.concatenate(items, axis=1)))
for key in result.keys():
if key.startswith('mosaic_'):
cv2.imshow('cv2: ' + key, tobytes(normalize(result[key])))
key = cv2.waitKey()
while key != ord('q') and key != ord('c'):
key = cv2.waitKey()
if key == ord('q'):
return False
return True
