def _seg(self, n_bkps=None, pen=None, epsilon=None):
"""Sequential peak search.
The stopping rule depends on the parameter passed to the function.
Args:
n_bkps (int): number of breakpoints to find before stopping.
penalty (float): penalty value (>0)
epsilon (float): reconstruction budget (>0)
Returns:
list: breakpoint index list
"""
# initialization
bkps = [self.n_samples]
stop = False
error = self.cost.sum_of_costs(bkps)
# peak search
peak_inds_shifted, = argrelmax(self.score,
order=max(self.width, self.min_size) //
(2 * self.jump),
mode="wrap")
if peak_inds_shifted.size == 0: # no peaks if the score is constant
return bkps
gains = np.take(self.score, peak_inds_shifted)
peak_inds_arr = np.take(self.inds, peak_inds_shifted)
# sort according to score value
_, peak_inds = unzip(sorted(zip(gains, peak_inds_arr)))
peak_inds = list(peak_inds)
while not stop:
stop = True
# _, bkp = max((v, k) for k, v in enumerate(self.score, start=1)
# if not any(abs(k - b) < self.width // 2 for b in bkps[:-1]))
try:
# index with maximum score
bkp = peak_inds.pop()
except IndexError: # peak_inds is empty
break
if n_bkps is not None:
if len(bkps) - 1 < n_bkps:
stop = False
elif pen is not None:
gain = error - self.cost.sum_of_costs(sorted([bkp] + bkps))
if gain > pen:
stop = False
elif epsilon is not None:
if error > epsilon:
stop = False
if not stop:
bkps.append(bkp)
bkps.sort()
error = self.cost.sum_of_costs(bkps)
return bkps
def _seg(self, n_bkps=None, pen=None, epsilon=None):
"""Sequential peak search.
The stopping rule depends on the parameter passed to the function.
Args:
n_bkps (int): number of breakpoints to find before stopping.
penalty (float): penalty value (>0)
epsilon (float): reconstruction budget (>0)
Returns:
list: breakpoint index list
"""
# initialization
bkps = [self.n_samples]
stop = False
error = self.cost.sum_of_costs(bkps)
# peak search
# forcing order to be above one in case jump is too large (issue #16)
order = max(max(self.width, 2 * self.min_size) // (2 * self.jump), 1)
peak_inds_shifted = argrelmax(self.score, order=order, mode="wrap")[0]
if peak_inds_shifted.size == 0: # no peaks if the score is constant
return bkps
gains = np.take(self.score, peak_inds_shifted)
peak_inds_arr = np.take(self.inds, peak_inds_shifted)
# sort according to score value
_, peak_inds = unzip(sorted(zip(gains, peak_inds_arr)))
peak_inds = list(peak_inds)
while not stop:
stop = True
# _, bkp = max((v, k) for k, v in enumerate(self.score, start=1)
# if not any(abs(k - b) < self.width // 2 for b in bkps[:-1]))
try:
# index with maximum score
bkp = peak_inds.pop()
except IndexError: # peak_inds is empty
break
if n_bkps is not None:
if len(bkps) - 1 < n_bkps:
stop = False
elif pen is not None:
gain = error - self.cost.sum_of_costs(sorted([bkp] + bkps))
if gain > pen:
stop = False
elif epsilon is not None:
if error > epsilon:
stop = False
if not stop:
bkps.append(bkp)
bkps.sort()
error = self.cost.sum_of_costs(bkps)
# remove shifts if they last too long
start_index = 0
end_index = 1
to_delete = []
if self.max_samples is not None and len(bkps) > 1:
for i in range(math.floor(len(bkps) / 2)):
if (bkps[end_index] - bkps[start_index]) > self.max_samples:
to_delete.append(start_index)
to_delete.append(end_index)
start_index += 2
end_index += 2
bkps = list(np.delete(bkps, to_delete))
# remove changepoint if it starts but never ends
if self.remove_single:
if (len(bkps) % 2) != 0:
del bkps[-1]
return bkps