def tracktorDeserialize(path, titles=None):
"""
get a track from the xml format from tracktor (1?)
"""
tree = ET.parse(path)
root = tree.getroot()
tracks = {}
for entry in root.find("COLLECTION").iter("ENTRY"):
track = Track()
track.name = entry.attrib["TITLE"]
track.path = entry.find("LOCATION").attrib["FILE"][:-4] #Removing .mp3
cues = [cue for cue in entry.iter("CUE_V2") if cue.attrib["NAME"] != "AutoGrid"]
track.features["Cues"] = Signal([cue.attrib["NAME"][:7] for cue in cues],
times=[float(cue.attrib["START"]) / 1000 for cue in cues],
sparse=True)
tracks[track.path] = track
if titles:
for t in titles:
if t in tracks:
yield tracks[t]
else:
print(t, "not in collection")
dummytrack = Track()
dummytrack.features["Cues"] = Signal(times=[])
yield dummytrack
# return [tracks[t] if t in tracks else Track() for t in titles]
else:
return tracks.values()
# bla = TraktorSerializer.tracktorDeserialize(
# "/home/mickael/Documents/programming/dj-tracks-switch-points/evaluation/mixed in key/collection.nml")
# print(bla)
def testPeakSelection(self):
# Basic test
myInput = [Signal(1, times=[0, 5, 10]), Signal(0.5, times=[0, 10])]
ps = PeakSelection()
result = ps.predictOne(myInput, None, None)[0]
self.assertEqual(list(result.values), [1.5, 1, 1.5])
self.assertEqual(list(result.times), [0, 5, 10])
# Test with mean
ps = PeakSelection(parameterMergeFunction=np.mean)
result = ps.predictOne(myInput, None, None)[0]
self.assertEqual(list(result.values), [0.75, 1, 0.75])
self.assertEqual(list(result.times), [0, 5, 10])
def testPeakPicking(self):
# Basic test
myInput = Signal([1.5, 1, 0, 0, 0, 2, 0], sampleRate=1)
pp = PeakPicking(parameterMinDistance=2, parameterRelativeThreshold=0.2)
result = pp.predictOne(myInput)[0]
self.assertEqual(list(result.values), [1.5, 2.])
self.assertEqual(list(result.times), [0, 5])
def testSignal(self):
# test sort of the values
signal = Signal([5, 4, 3, 2, 1], times=[5, 4, 3, 2, 1])
self.assertEqual(list(signal.values), [1, 2, 3, 4, 5])
# test assertion of duplicate values
# with self.assertRaises(AssertionError):
signal = Signal([5, 4, 3, 2, 1], times=[5, 4, 3, 3, 1])
# test Qantization: remove doubles
signal = Signal(1, times=[0, 1.1, 2.1, 2.5, 2.6, 2.7, 3.1])
grid = Signal(-1, times=list(range(5)))
signal.quantizeTo(grid)
# In case 1 value is exactly between 2 grid ticks, use the smallest one
self.assertEqual(list(signal.values), [1, 1, 2, 3])
# test Quantization: maxThreshold for out of bound
signal = Signal(1, times=[0, 1.1, 2.0, 2.1, 2.6, 2.7, 3.1])
signal.quantizeTo(grid, maxThreshold=0.2)
self.assertEqual(list(signal.times), [0, 1, 2, 3])
self.assertEqual(list(signal.values), [1, 1, 2, 1])
# test quantization: don't remove duplicates
signal = Signal(1, times=[0, 1.1, 2.0, 2.1, 2.6, 2.7, 3.1])
signal.quantizeTo(grid, maxThreshold=0.2, removeDuplicatedValues=False)
self.assertEqual(list(signal.times), [0, 1, 2, 2, 3])
self.assertEqual(list(signal.values), [1, 1, 1, 1, 1])
# test quantization: don't remove out of bounds
signal = Signal(1, times=[0, 1.1, 2.0, 2.1, 2.6, 2.7, 3.1])
signal.quantizeTo(grid, maxThreshold=0.2, removeDuplicatedValues=False, removeOutOfBound=False)
self.assertEqual(list(signal.times), [0, 1, 2, 2, 2.6, 2.7, 3])
self.assertEqual(list(signal.values), [1, 1, 1, 1, 1, 1, 1])
# test getIndex
signal = Signal(1, times=[0, 1.1, 2.0, 2.1, 2.6, 2.7, 3.1])
idx = signal.getIndex(2.2, toleranceWindow=0.5)
self.assertEqual(idx, 3)
# test clusterSignals
result = Signal.clusterSignals([SparseSignal(1, [0, 1, 2, 3]), SparseSignal(0, [1, 3, 3.1])], minDistance=0, mergeValue=np.mean)
self.assertEqual(list(result.times), [0, 1, 2, 3, 3.1])
self.assertEqual(list(result.values), [1, 0.5, 1, .5, 0])
def testStructureEval(self):
# basic tests
result = evalCuesMutliple([Signal(1, times=[1, 3, 2]), Signal(1, times=[1, 2, 3])],
[Signal(1, times=[1, 2, 3]), Signal(1, times=[1, 2])])
self.assertEqual(result["recall"], 1)
self.assertEqual(result["precision"], 5 / 6)
self.assertEqual(result["fMeasure"], hmean([1, 5 / 6]))
result = evalCuesMutliple([Signal(1, times=[1, 2, 4, 5, 6])], [Signal(1, times=[1, 2, 3, 4])], limitSearchSpace=True)
self.assertEqual(result["recall"], 3 / 4)
self.assertEqual(result["precision"], 3 / 3)
result = evalCuesMutliple([Signal(1, times=[1, 2, 5, 6]), Signal(1, times=[1, 2, 3, 4])],
[Signal(1, times=[1, 2, 3, 4]), Signal(1, times=[1, 2, 5, 6])],
averagePerDocument=True,
returnDetails=True,
limitSearchSpace=True)
self.assertEqual(result["recall"], [0.5, 0.5])
self.assertEqual(result["precision"], [1, 0.5])
self.assertEqual(result["fMeasure"], list(hmean([[0.5, 0.5], [1, 0.5]], axis=0)))
# test duration
result = evalCuesMutliple([Signal(1, times=[1, 2, 3])], [SparseSegmentSignal(1, [[1, 2], [1.6, 2.2]])])
self.assertEqual(result["recall"], 1)
self.assertEqual(result["precision"], 2 / 3)
def testWindow(self):
# RMS value normalize by sample number
w = Windowing()
input = Signal(2, times=list(range(20)))
result, = w.predictOne(input, Signal(0, times=[-1, 2.1, 3.1, 4.1, 20]))
self.assertEqual(result[0], result[2])