def multiChorusSections(intvs, dur):
"""avoid intersection of tuned intervals"""
def key(x):
# timestamp with precision of 0.01s
return int(x * 100)
# value 1=chorus begin -1=chorus end
boundaries = defaultdict(int)
for intv in intvs:
boundaries[key(intv[0])] += 1
boundaries[key(intv[1])] -= 1
intervals, labels = [[0, 0]], [CLF_NON_TARGET_LABEL]
state = 0 # 0:others >0:chorus
for bdr in sorted(boundaries.keys()):
t = bdr / 100.0
intervals[-1][1] = t
intervals.append([t, 0])
state += boundaries[bdr]
if state == 0:
labels.append(CLF_NON_TARGET_LABEL)
elif state > 0:
labels.append(CLF_TARGET_LABEL)
else:
logger.error(f"invalid state, boundaries={boundaries}")
intervals[-1][1] = dur
mirexFmt = (np.array(intervals), np.array(labels, dtype="U16"))
logger.debug(f"multi chorus sections, output=\n{mirexLines(mirexFmt)}")
return mergeIntervals(mirexFmt)
def loadFeature(self, i):
# <ddir>/<orig_name>-<id>.pkl
pklPath = self.getPklPath(i)
try:
with open(pklPath, "rb") as f:
feature = pickle.load(f)
return feature
except FileNotFoundError as e:
logger.error(
f'file "{pklPath}" not found, build the dataset first.')
raise e
def __call__(self):
try:
with Pool(self.num_workers) as p:
N = len(self.dataset)
metrics = list(tqdm(p.imap(self.eval, range(N)), total=N))
except RuntimeError as e:
# CUDA RuntimeError
metrics = [self.eval(i) for i in tqdm(range(N))]
logger.error(f"[RuntimeError] ", e)
titles = [sample["title"] for sample in self.dataset]
return np.array(metrics), titles
def getResult(self, algoName, titles=None):
try:
idx = self.algoNames.index(algoName)
except ValueError:
logger.error(f"{algoName} results not found in {self.datasetName}")
return None
metrics = self.metricsList[idx]
_titles = self.titlesList[idx]
titles = _titles if titles is None else titles
res = [metrics[_titles.index(title)] for title in titles]
return np.array(res), titles
def loadData(self, dataFile):
if os.path.exists(dataFile):
with open(dataFile, "rb") as f:
X, y = pickle.load(f)
logger.info(
f"<{self.__class__.__name__}> load data from '{dataFile}'")
logger.info(
f"target(chorus)/total={sum(np.array(y)==CLF_TARGET_LABEL)}/{len(y)}"
)
else:
logger.error(f"build dataset for classifier first")
raise FileNotFoundError(dataFile)
return X, y