def sample_gp(length, sigma, n=1, noise=0.001):
""" sample a function from a Gaussian process with Gaussian kernel """
X = np.linspace(0, 1, length, False).reshape([1, length])
zeroY = np.zeros(length)
K = maxdiv_util.calc_gaussian_kernel(
X, sigma / length) + noise * np.eye(X.shape[1])
return np.random.multivariate_normal(zeroY, K, n)
def sample_gp(length, dim=1, sigma=0.02, noise=0.001):
""" sample a function from a Gaussian process with Gaussian kernel """
X = np.arange(0, length / 250.0, 0.004)
X = np.reshape(X, [1, len(X)])
meany = np.zeros(X.shape[1])
K = maxdiv_util.calc_gaussian_kernel(X, sigma) + noise * np.eye(X.shape[1])
return np.random.multivariate_normal(meany, K, dim)
def sample_gp(X, meany, sigma, n=1, noise=0.001):
""" sample a function from a Gaussian process with Gaussian kernel """
K = maxdiv_util.calc_gaussian_kernel(X, sigma) + noise * np.eye(X.shape[1])
return np.random.multivariate_normal(meany, K, n)
# Add a constant offset to avoid negative scores
base_score = min(score for _, _, score in norm_scores) - 0.01
for i, (a, b, score) in enumerate(norm_scores):
norm_scores[i] = (a, b, score - base_score)
scores[meth] = norm_scores
elif meth == 'gaussian_cov_ts':
scores[meth] = maxdiv.maxdiv_gaussian(pts, proposals, mode = 'TS', gaussian_mode = 'COV')
# de-normalize
chi_mean = (pts.shape[0] * (pts.shape[0] + 3)) / 2
chi_sd = np.sqrt(2 * chi_mean)
for i, (a, b, score) in enumerate(scores[meth]):
scores[meth][i] = (a, b, score * chi_sd + chi_mean)
elif meth.startswith('gaussian'):
scores[meth] = maxdiv.maxdiv_gaussian(pts, proposals, mode = 'I_OMEGA', gaussian_mode = meth[9:].upper())
elif meth == 'parzen':
K = maxdiv_util.calc_gaussian_kernel(pts)
scores[meth] = maxdiv.maxdiv_parzen(K, proposals, mode = 'I_OMEGA')
else:
print('Unknown method: {}'.format(meth))
exit()
if method != 'compare':
# Plot top 5 detections
detections = maxdiv.find_max_regions(scores[method])
eval.plotDetections(ts, detections[:5])
# Plot histogram of the length of detected intervals
plt.title('Histogram of the Length of Detected Intervals for Method "{}"'.format(method))
plt.hist([b - a for a, b, score in detections])
plt.show()