def main(tseries_fpath, k, plot_foldpath):
import mkl
mkl.set_num_threads(16)
initialize_matplotlib()
X = np.genfromtxt(tseries_fpath)[:, 1:]
aux = X.sum(axis=1)
fix = np.where(aux == 0)[0]
X[fix] += .001 #fixing zero only rows
X = X.copy()
cent, assign, shift, dists_cent = ksc.inc_ksc(X, k)
for i in range(cent.shape[0]):
t_series = cent[i]
plt.plot(t_series, '-k')
plt.gca().get_xaxis().set_visible(False)
plt.gca().get_yaxis().set_visible(False)
#plt.ylabel('Views')
#plt.xlabel('Time')
plt.savefig(os.path.join(plot_foldpath, '%d.pdf' % i))
plt.close()
half = t_series.shape[0] // 2
to_shift = half - np.argmax(t_series)
to_plot_peak_center = dist.shift(t_series, to_shift, rolling=True)
plt.plot(to_plot_peak_center, '-k')
plt.gca().get_xaxis().set_visible(False)
plt.gca().get_yaxis().set_visible(False)
#plt.ylabel('Views')
#plt.xlabel('Time')
plt.savefig(os.path.join(plot_foldpath, '%d-peak-center.pdf' % i))
plt.close()
to_shift = 0 - np.argmin(t_series)
to_plot_min_first = dist.shift(t_series, to_shift, rolling=True)
plt.plot(to_plot_min_first, '-k')
plt.gca().get_xaxis().set_visible(False)
plt.gca().get_yaxis().set_visible(False)
#plt.ylabel('Views')
#plt.xlabel('Time')
plt.savefig(os.path.join(plot_foldpath, '%d-min-first.pdf' % i))
plt.close()
np.savetxt(os.path.join(plot_foldpath, 'cents.dat'), cent, fmt='%.5f')
np.savetxt(os.path.join(plot_foldpath, 'assign.dat'), assign, fmt='%d')
np.savetxt(os.path.join(plot_foldpath, 'shift.dat'), shift, fmt='%d')
np.savetxt(os.path.join(plot_foldpath, 'dists_cent.dat'),
dists_cent,
fmt='%.5f')
def plot_series(t_series, plot_foldpath, name, shift=False):
to_plot = t_series
if shift:
to_shift = 0 - np.argmin(t_series)
to_plot = dist.shift(t_series, to_shift, rolling=True)
plt.plot(to_plot, '-k')
plt.ylabel('Views')
plt.xlabel('Time')
plt.savefig(os.path.join(plot_foldpath, '%s.png' % name))
plt.close()
def plot_series(t_series, plot_foldpath, name, shift=False):
to_plot = t_series
if shift:
to_shift = 0 - np.argmin(t_series)
to_plot = dist.shift(t_series, to_shift, rolling=True)
plt.plot(to_plot, '-k')
plt.ylabel('Views')
plt.xlabel('Time')
plt.savefig(os.path.join(plot_foldpath, '%s.png' % name))
plt.close()
def _compute_centroids(tseries, assign, num_clusters, to_shift=None):
'''
Given a time series matrix and cluster assignments, this method will
compute the spectral centroids for each cluster.
Arguments
---------
tseries: matrix (n_series, n_points)
Time series beng clustered
assign: array of ints (size = n_series)
The cluster assignment for each time series
num_clusters: int
The number of clusters being searched for
to_shift (optional): array of ints (size = n_series)
Determines if time series should be shifted, if different from `None`.
In this case, each series will be shifted by the corresponding amount
in the array.
'''
series_size = tseries.shape[1]
centroids = np.ndarray((num_clusters, series_size))
#shift series for best centroid distance
#TODO: this method can be cythonized and done in parallel
shifted = tseries
if to_shift is not None:
for i in xrange(tseries.shape[0]):
shifted[i] = shift(tseries[i], to_shift[i], rolling=True)
#compute centroids
for k in xrange(num_clusters):
members = shifted[assign == k]
if members.any():
num_members = 0
if members.ndim == 2:
axis = 1
num_members = members.shape[0]
else:
axis = 0
num_members = 1
ssqs = np.tile(np.sum(members**2, axis=axis), (series_size, 1))
#the original papers divides by ssqs only, while the author's
#example implementation uses sqrt. We chose sqrt because it appears
#to yield better centroids.
aux = members / np.sqrt(ssqs.T)
x_mat = np.dot(aux.T, aux)
i_mat = num_members * np.eye(series_size)
m_mat = i_mat - x_mat
#compute eigenvalues and chose the vector for the smallest one
#TODO: Check if using scipy's linalg is faster (has more options
# such as finding only the smallest eigval)
_, eig_vectors = LA.eigh(m_mat, eigvals=(0, 0))
centroids[k] = eig_vectors[:,0]
if centroids[k].sum() < 0:
centroids[k] = -centroids[k]
else:
centroids[k] = np.zeros(series_size)
return centroids
def test_shift_drop(self):
array = np.array([1.0])
assert_array_equal(np.array([1.0]), dist.shift(array, 0, False))
assert_array_equal(np.array([0.0]), dist.shift(array, 1, False))
assert_array_equal(np.array([0.0]), dist.shift(array, 1, False))
assert_array_equal(np.array([0.0]), dist.shift(array, -2, False))
assert_array_equal(np.array([0.0]), dist.shift(array, -2, False))
array = np.array([1.0, 2.0, 3.0, 4.0])
assert_array_equal(np.array([1.0, 2.0, 3.0, 4.0]), dist.shift(array, 0, False))
assert_array_equal(np.array([0.0, 1.0, 2.0, 3.0]), dist.shift(array, 1, False))
assert_array_equal(np.array([2.0, 3.0, 4.0, 0.0]), dist.shift(array, -1, False))
assert_array_equal(np.array([0.0, 0.0, 1.0, 2.0]), dist.shift(array, 2, False))
assert_array_equal(np.array([3.0, 4.0, 0.0, 0.0]), dist.shift(array, -2, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 1.0]), dist.shift(array, 3, False))
assert_array_equal(np.array([4.0, 0.0, 0.0, 0.0]), dist.shift(array, -3, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 0.0]), dist.shift(array, 4, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 0.0]), dist.shift(array, -4, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 0.0]), dist.shift(array, 5, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 0.0]), dist.shift(array, -5, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 0.0]), dist.shift(array, 50, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 0.0]), dist.shift(array, -50, False))
def test_shift_roll(self):
array = np.array([])
assert_array_equal(np.array([]), dist.shift(array, 0))
assert_array_equal(np.array([]), dist.shift(array, -1))
assert_array_equal(np.array([]), dist.shift(array, 1))
assert_array_equal(np.array([]), dist.shift(array, 10))
assert_array_equal(np.array([]), dist.shift(array, -10))
array = np.array([1.0])
assert_array_equal(np.array([1.0]), dist.shift(array, 0, True))
assert_array_equal(np.array([1.0]), dist.shift(array, 1, True))
assert_array_equal(np.array([1.0]), dist.shift(array, 1, True))
assert_array_equal(np.array([1.0]), dist.shift(array, -2, True))
assert_array_equal(np.array([1.0]), dist.shift(array, -2, True))
array = np.array([1.0, 2.0, 3.0, 4.0])
assert_array_equal(np.array([1.0, 2.0, 3.0, 4.0]), dist.shift(array, 0, True))
assert_array_equal(np.array([4.0, 1.0, 2.0, 3.0]), dist.shift(array, 1, True))
assert_array_equal(np.array([2.0, 3.0, 4.0, 1.0]), dist.shift(array, -1, True))
assert_array_equal(np.array([3.0, 4.0, 1.0, 2.0]), dist.shift(array, 2, True))
assert_array_equal(np.array([3.0, 4.0, 1.0, 2.0]), dist.shift(array, -2, True))
assert_array_equal(np.array([2.0, 3.0, 4.0, 1.0]), dist.shift(array, 3, True))
assert_array_equal(np.array([4.0, 1.0, 2.0, 3.0]), dist.shift(array, -3, True))
assert_array_equal(np.array([1.0, 2.0, 3.0, 4.0]), dist.shift(array, 4, True))
assert_array_equal(np.array([1.0, 2.0, 3.0, 4.0]), dist.shift(array, -4, True))
assert_array_equal(np.array([4.0, 1.0, 2.0, 3.0]), dist.shift(array, 5, True))
assert_array_equal(np.array([2.0, 3.0, 4.0, 1.0]), dist.shift(array, -5, True))
assert_array_equal(np.array([1.0, 2.0, 3.0, 4.0]), dist.shift(array, 8, True))
assert_array_equal(np.array([1.0, 2.0, 3.0, 4.0]), dist.shift(array, -8, True))
def _compute_centroids(tseries, assign, num_clusters, to_shift=None):
'''
Given a time series matrix and cluster assignments, this method will
compute the spectral centroids for each cluster.
Arguments
---------
tseries: matrix (n_series, n_points)
Time series beng clustered
assign: array of ints (size = n_series)
The cluster assignment for each time series
num_clusters: int
The number of clusters being searched for
to_shift (optional): array of ints (size = n_series)
Determines if time series should be shifted, if different from `None`.
In this case, each series will be shifted by the corresponding amount
in the array.
'''
series_size = tseries.shape[1]
centroids = np.ndarray((num_clusters, series_size))
#shift series for best centroid distance
#TODO: this method can be cythonized and done in parallel
shifted = tseries
if to_shift is not None:
for i in range(tseries.shape[0]):
shifted[i] = shift(tseries[i], to_shift[i], rolling=True)
#compute centroids
for k in range(num_clusters):
members = shifted[assign == k]
if members.any():
num_members = 0
if members.ndim == 2:
axis = 1
num_members = members.shape[0]
else:
axis = 0
num_members = 1
ssqs = np.tile(np.sum(members**2, axis=axis), (series_size, 1))
#the original papers divides by ssqs only, while the author's
#example implementation uses sqrt. We chose sqrt because it appears
#to yield better centroids.
aux = members / np.sqrt(ssqs.T)
x_mat = np.dot(aux.T, aux)
i_mat = num_members * np.eye(series_size)
m_mat = i_mat - x_mat
#compute eigenvalues and chose the vector for the smallest one
#TODO: Check if using scipy's linalg is faster (has more options
# such as finding only the smallest eigval)
_, eig_vectors = LA.eigh(m_mat, eigvals=(0, 0))
centroids[k] = eig_vectors[:, 0]
if centroids[k].sum() < 0:
centroids[k] = -centroids[k]
else:
centroids[k] = np.zeros(series_size)
return centroids
def test_shift_drop(self):
array = np.array([1.0])
assert_array_equal(np.array([1.0]), dist.shift(array, 0, False))
assert_array_equal(np.array([0.0]), dist.shift(array, 1, False))
assert_array_equal(np.array([0.0]), dist.shift(array, 1, False))
assert_array_equal(np.array([0.0]), dist.shift(array, -2, False))
assert_array_equal(np.array([0.0]), dist.shift(array, -2, False))
array = np.array([1.0, 2.0, 3.0, 4.0])
assert_array_equal(np.array([1.0, 2.0, 3.0, 4.0]),
dist.shift(array, 0, False))
assert_array_equal(np.array([0.0, 1.0, 2.0, 3.0]),
dist.shift(array, 1, False))
assert_array_equal(np.array([2.0, 3.0, 4.0, 0.0]),
dist.shift(array, -1, False))
assert_array_equal(np.array([0.0, 0.0, 1.0, 2.0]),
dist.shift(array, 2, False))
assert_array_equal(np.array([3.0, 4.0, 0.0, 0.0]),
dist.shift(array, -2, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 1.0]),
dist.shift(array, 3, False))
assert_array_equal(np.array([4.0, 0.0, 0.0, 0.0]),
dist.shift(array, -3, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 0.0]),
dist.shift(array, 4, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 0.0]),
dist.shift(array, -4, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 0.0]),
dist.shift(array, 5, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 0.0]),
dist.shift(array, -5, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 0.0]),
dist.shift(array, 50, False))
assert_array_equal(np.array([0.0, 0.0, 0.0, 0.0]),
dist.shift(array, -50, False))
def test_shift_roll(self):
array = np.array([])
assert_array_equal(np.array([]), dist.shift(array, 0))
assert_array_equal(np.array([]), dist.shift(array, -1))
assert_array_equal(np.array([]), dist.shift(array, 1))
assert_array_equal(np.array([]), dist.shift(array, 10))
assert_array_equal(np.array([]), dist.shift(array, -10))
array = np.array([1.0])
assert_array_equal(np.array([1.0]), dist.shift(array, 0, True))
assert_array_equal(np.array([1.0]), dist.shift(array, 1, True))
assert_array_equal(np.array([1.0]), dist.shift(array, 1, True))
assert_array_equal(np.array([1.0]), dist.shift(array, -2, True))
assert_array_equal(np.array([1.0]), dist.shift(array, -2, True))
array = np.array([1.0, 2.0, 3.0, 4.0])
assert_array_equal(np.array([1.0, 2.0, 3.0, 4.0]),
dist.shift(array, 0, True))
assert_array_equal(np.array([4.0, 1.0, 2.0, 3.0]),
dist.shift(array, 1, True))
assert_array_equal(np.array([2.0, 3.0, 4.0, 1.]),
dist.shift(array, -1, True))
assert_array_equal(np.array([3.0, 4.0, 1.0, 2.0]),
dist.shift(array, 2, True))
assert_array_equal(np.array([3.0, 4.0, 1.0, 2.0]),
dist.shift(array, -2, True))
assert_array_equal(np.array([2.0, 3.0, 4.0, 1.0]),
dist.shift(array, 3, True))
assert_array_equal(np.array([4.0, 1.0, 2.0, 3.0]),
dist.shift(array, -3, True))
assert_array_equal(np.array([1.0, 2.0, 3.0, 4.0]),
dist.shift(array, 4, True))
assert_array_equal(np.array([1.0, 2.0, 3.0, 4.0]),
dist.shift(array, -4, True))
assert_array_equal(np.array([4.0, 1.0, 2.0, 3.0]),
dist.shift(array, 5, True))
assert_array_equal(np.array([2.0, 3.0, 4.0, 1.]),
dist.shift(array, -5, True))
assert_array_equal(np.array([1.0, 2.0, 3.0, 4.0]),
dist.shift(array, 8, True))
assert_array_equal(np.array([1.0, 2.0, 3.0, 4.0]),
dist.shift(array, -8, True))
