def cross_validation(T, y):
"""Use Cross Validation (leave-one-out) to select features.
Args:
T: feature statistics list
y: labels
"""
from sklearn.model_selection import LeaveOneOut
y = np.array(y)
judge = list()
for train_index, valid_index in LeaveOneOut().split(T):
T_train = T[train_index]
T_valid = T[valid_index]
y_train = y[train_index]
y_valid = y[valid_index]
T_train, mean, std = feature.normalize(T_train)
T_principle, T_principle_index, dist, AUC = feature_select(T_train,
y_train, k=3)
ts = threshold(dist, y_train)
C = gen_center(T_principle, y_train)
T_valid = (T_valid - mean) / std
dist_valid = util.distance(T_valid.T[T_principle_index].T, C)
if y_valid[0] == 1:
if dist_valid[0] < ts:
judge.append(1)
else:
judge.append(0)
else:
if dist_valid[0] < ts:
judge.append(0)
else:
judge.append(1)
accuracy = sum(judge) / len(judge)
return accuracy
def cross_validation2(T, y):
"""Use Cross Validation (leave-one-out) to select features.
Args:
T: feature statistics list
y: labels
"""
from sklearn.model_selection import LeaveOneOut
y = np.array(y)
judge = list()
T_principle_index = np.array([0, 18, 43])
for train_index, valid_index in LeaveOneOut().split(T):
T_train = T[train_index]
T_valid = T[valid_index]
y_train = y[train_index]
T_train, mean, std = feature.normalize(T_train)
T_principle = T_train.T[T_principle_index].T
C = gen_center(T_principle, y_train)
dist = util.distance(T_principle, C)
ts = threshold(dist, y_train)
T_valid = (T_valid - mean) / std
dist_valid = util.distance(T_valid.T[T_principle_index].T, C)
if dist_valid[0] < ts:
judge.append(1)
else:
judge.append(0)
return np.array(judge)
def guess_syllables(wave, window_size=10, len_threshold=0.01):
amp, _ = segmentation(wave)
windowed_amp = feature.normalize(moving_window(amp, window_size))
maximi=signal.argrelmax(windowed_amp, order=7)
#plot(windowed_amp)
#show()
return len(maximi[0])
def plotstuff(filename):
a=fileio.Wave("audio/samples/"+filename+".wav")
am, fr = syllable.segmentation(a)
title(filename)
plot(feature.normalize(syllable.moving_window(am, 10)), label="amp")
#plot(feature.normalize(syllable.moving_window(fr, 10)), label="freq")
#plot(feature.normalize(am), label="amp")
#plot(feature.normalize(fr), label="freq")
legend(framealpha=0.5);
show()
def populateSampleData(sample_directory, freq=False):
print("loading data")
files = readAllFilesInDirectory(sample_directory);
sampleData = {}
abstractSampleData={}
fileNames = set([x[1] for x in files])
for name in fileNames:
sampleData[name] = []
abstractSampleData[name] =[]
for file in files:
sampleData[file[1]].append(file[0])
#compute signatures:
for key in sampleData.keys():
processed=[]
for sample in sampleData[key]:
processed.append(feature.normalize(feature.abstract_cartoon(sample, freq=freq)))
print(key, len(processed))
for i in range(1,len(processed)):
processed[i] = feature.align_peaks(processed[i], processed[i-1], 1)+processed[i-1]
#abstractSampleData[key] = feature.normalize(feature.align_peaks(processed[0], processed[1], 1) + processed[1])
abstractSampleData[key] = feature.normalize(processed[-1])
return abstractSampleData
def predict(T, y, T_test):
y = np.array(y)
T, mean, std = feature.normalize(T)
T_principle, T_principle_index, dist, AUC = feature_select(T, y, k=4)
ts = threshold(dist, y)
C = gen_center(T_principle, y)
T_test = (T_test - mean) / std
dist_test = util.distance(T_test.T[T_principle_index].T, C)
judge = list()
for d in dist_test:
if d < ts:
judge.append(1)
else:
judge.append(0)
return np.array(judge), T_principle_index, AUC, dist