def pw_wavy(n_samples=200, n_bkps=3, noise_std=None):
"""Return a 1D piecewise wavy signal and the associated changepoints.
Args:
n_samples (int, optional): signal length
n_bkps (int, optional): number of changepoints
noise_std (float, optional): noise std. If None, no noise is added
Returns:
tuple: signal of shape (n_samples, 1), list of breakpoints
"""
# breakpoints
bkps = draw_bkps(n_samples, n_bkps)
# we create the signal
f1 = np.array([0.075, 0.1])
f2 = np.array([0.1, 0.125])
freqs = np.zeros((n_samples, 2))
for sub, val in zip(np.split(freqs, bkps[:-1]), cycle([f1, f2])):
sub += val
tt = np.arange(n_samples)
signal = np.sum((np.sin(2 * np.pi * tt * f) for f in freqs.T))
if noise_std is not None:
noise = normal(scale=noise_std, size=signal.shape)
signal += noise
return signal, bkps
def pw_constant(n_samples=200, n_features=1, n_bkps=3, noise_std=None, delta=(1, 10)):
"""Return a piecewise constant signal and the associated changepoints.
Args:
n_samples (int): signal length
n_features (int, optional): number of dimensions
n_bkps (int, optional): number of changepoints
noise_std (float, optional): noise std. If None, no noise is added
delta (tuple, optional): (delta_min, delta_max) max and min jump values
Returns:
tuple: signal of shape (n_samples, n_features), list of breakpoints
"""
# breakpoints
bkps = draw_bkps(n_samples, n_bkps)
# we create the signal
signal = np.empty((n_samples, n_features), dtype=float)
tt_ = np.arange(n_samples)
delta_min, delta_max = delta
# mean value
center = np.zeros(n_features)
for ind in np.split(tt_, bkps):
if ind.size > 0:
# jump value
jump = rd.uniform(delta_min, delta_max, size=n_features)
spin = rd.choice([-1, 1], n_features)
center += jump * spin
signal[ind] = center
if noise_std is not None:
noise = rd.normal(size=signal.shape) * noise_std
signal = signal + noise
return signal, bkps
def pw_normal(n_samples=200, n_bkps=3):
"""Return a 2D piecewise Gaussian signal and the associated changepoints.
Args:
n_samples (int, optional): signal length
n_bkps (int, optional): number of change points
Returns:
tuple: signal of shape (n_samples, 2), list of breakpoints
"""
# breakpoints
bkps = draw_bkps(n_samples, n_bkps)
# we create the signal
signal = np.zeros((n_samples, 2), dtype=float)
cov1 = np.array([[1, 0.9], [0.9, 1]])
cov2 = np.array([[1, -0.9], [-0.9, 1]])
for sub, cov in zip(np.split(signal, bkps), cycle((cov1, cov2))):
n_sub, _ = sub.shape
sub += rd.multivariate_normal([0, 0], cov, size=n_sub)
return signal, bkps
def pw_constant(n_samples=200,
n_features=1,
n_bkps=3,
noise_std=None,
delta=(1, 10),
noise_delta=(1, 1)):
"""Return a piecewise constant signal and the associated changepoints.
Args:
n_samples (int): signal length
n_features (int, optional): number of dimensions
n_bkps (int, optional): number of changepoints
noise_std (float, optional): noise std. If None or 0, no noise is added
delta (tuple, optional): (delta_min, delta_max) max and min jump values
noise_delta (tuple, optional): (noise_min_mult, noise_max_mult) max and min noise multipliers
Returns:
tuple: signal of shape (n_samples, n_features), list of breakpoints
noise_min_mult and noise_max_mult must be greater than 0
0 to 1 and 1 to inf is equaly distributed.
"""
if noise_std is None:
noise_std = 0
# breakpoints
bkps = draw_bkps(n_samples, n_bkps)
# we create the signal
signal = np.empty((n_samples, n_features), dtype=float)
signal_noise = np.empty((n_samples, n_features), dtype=float)
tt_ = np.arange(n_samples)
delta_min, delta_max = delta
noise_delta_min, noise_delta_max = noise_delta
if not noise_delta_min <= noise_delta_max:
sys.exit('Error: noise_delta_min is bigger than noise_delta_max')
# mean value
center = np.zeros(n_features)
for ind in np.split(tt_, bkps):
if ind.size > 0:
# jump value
jump = rd.uniform(delta_min, delta_max, size=n_features)
spin = rd.choice([-1, 1], n_features)
center += jump * spin
signal[ind] = center
noise_center = np.ones(n_features)
noise_jump = np.zeros(n_features)
for ind in np.split(tt_, bkps):
if ind.size > 0:
if noise_delta_min >= 1 and noise_delta_max >= 1:
noise_jump = rd.uniform(noise_delta_min, noise_delta_max,
n_features)
elif noise_delta_min < 1 and noise_delta_max <= 1:
aux_jump = rd.uniform(1 / noise_delta_min, 1 / noise_delta_max,
n_features)
noise_jump = 1 / aux_jump
else: #equalize probabilities lower than 1 and bigger than 1 taking into account the proportion
inv_min = 1 / noise_delta_min
domain_len = inv_min + noise_delta_max
p = [inv_min / domain_len, noise_delta_max / domain_len]
noise_side = rd.choice([-1, 1], n_features, p=p)
# print('ns',noise_side)
for f in range(n_features):
if noise_side[f] == 1:
noise_jump[f] = rd.choice([1, noise_delta_max])
# print('!!!!',noise_jump[f])
else:
aux_jump = rd.choice([inv_min, 1])
noise_jump[f] = 1 / aux_jump
# print('nc1',noise_center)
# print('nj1',noise_jump)
noise_center *= noise_jump
# print('nc2',noise_center)
signal_noise[ind] = noise_center
noise = rd.normal(size=signal.shape) * signal_noise * noise_std
signal = signal + noise
return signal, bkps