def pretty_print_gmm(gmm):
from gmm_helper import group_gmm_param_from_gmm_param_array
if not isinstance(gmm[0], np.ndarray) and not isinstance(gmm[0], list):
gmm = group_gmm_param_from_gmm_param_array(gmm, sort_group=True)
pretty_result = pd.DataFrame(
gmm, columns=['weight', 'mean_x', 'mean_y', 'sig_x', 'sig_y', 'corr'])
pretty_result.index += 1
decimal_format = lambda x: "{0:.3f}".format(x)
return pretty_result.applymap(decimal_format)
def pretty_print_gmm(gmm):
from gmm_helper import group_gmm_param_from_gmm_param_array
if not isinstance(gmm[0], np.ndarray) and not isinstance(gmm[0], list):
gmm = group_gmm_param_from_gmm_param_array(gmm, sort_group=True)
pretty_result = pd.DataFrame(gmm, columns=['weight',
'mean_x', 'mean_y',
'sig_x', 'sig_y','corr'])
pretty_result.index += 1
decimal_format = lambda x: "{0:.3f}".format(x)
return pretty_result.applymap(decimal_format)
def plot_gmm_ellipses(gmm, ax=None, xlabel='x', ylabel='y'):
from operator import itemgetter
if 'sns' in globals():
prop_cycle = iter(sns.color_palette("hls", 6))
else:
prop_cycle = iter(mpl.rcParams['axes.color_cycle'])
if ax is None:
fig, ax = plt.subplots(figsize=(3.5, 3.5))
print 'GMM Plot Result'
if not isinstance(gmm[0], np.ndarray) and not isinstance(gmm[0], list):
gmm = group_gmm_param_from_gmm_param_array(gmm, sort_group=False)
gmm = sorted(gmm, key=itemgetter(0), reverse=True)
for i, g in enumerate(gmm):
xy_mean = np.matrix([g[1], g[2]])
sigx, sigy, sigxy = g[3], g[4], g[5] * g[3] * g[4]
cov_matrix = np.matrix([[sigx**2, sigxy], [sigxy, sigy**2]])
# eigenvalues, and eigen vector
w, v = np.linalg.eigh(cov_matrix)
# normalized the eigen vector
uu = v[0] / np.linalg.norm(v[0])
# rotation angle, from x, y to u, v
angle_arc = np.arctan2(uu[0, 1], uu[0, 0])
angle = np.degrees(angle_arc)
transform_matrix = np.matrix([[np.cos(angle_arc), -np.sin(angle_arc)],
[np.sin(angle_arc),
np.cos(angle_arc)]])
xy_mean_in_uv = transform_matrix * xy_mean.T
# print fraction, rotation agnle, u v mean(in standalone panel), std
print g[0], xy_mean, np.sqrt(w), angle
ell = mpl.patches.Ellipse(xy=xy_mean.T,
width=2 * np.sqrt(w[0]),
height=2 * np.sqrt(w[1]),
angle=angle,
alpha=0.7,
color=next(prop_cycle),
label="{0:.3f}".format(g[0]))
ax.add_patch(ell)
ax.autoscale()
ax.set_aspect('equal')
plt_configure(xlabel='x', ylabel='y', legend={'loc': 'best'})
def plot_gmm_ellipses(gmm, ax=None, xlabel='x', ylabel='y'):
from operator import itemgetter
if 'sns' in globals():
prop_cycle = iter(sns.color_palette("hls", 6))
else:
prop_cycle = iter(mpl.rcParams['axes.color_cycle'])
if ax is None:
fig, ax = plt.subplots(figsize=(3.5, 3.5))
print 'GMM Plot Result'
if not isinstance(gmm[0], np.ndarray) and not isinstance(gmm[0], list):
gmm = group_gmm_param_from_gmm_param_array(gmm, sort_group=False)
gmm = sorted(gmm, key=itemgetter(0),reverse=True)
for i, g in enumerate(gmm):
xy_mean = np.matrix([g[1], g[2]])
sigx, sigy, sigxy = g[3], g[4], g[5]*g[3]*g[4]
cov_matrix = np.matrix([[sigx**2, sigxy], [sigxy, sigy**2]])
# eigenvalues, and eigen vector
w, v = np.linalg.eigh(cov_matrix)
# normalized the eigen vector
uu = v[0] / np.linalg.norm(v[0])
# rotation angle, from x, y to u, v
angle_arc = np.arctan2(uu[0, 1], uu[0, 0])
angle = np.degrees(angle_arc)
transform_matrix = np.matrix([[np.cos(angle_arc), -np.sin(angle_arc)],
[np.sin(angle_arc), np.cos(angle_arc)]])
xy_mean_in_uv = transform_matrix * xy_mean.T
# print fraction, rotation agnle, u v mean(in standalone panel), std
print g[0], xy_mean, np.sqrt(w), angle
ell = mpl.patches.Ellipse(xy=xy_mean.T, width=2*np.sqrt(w[0]), height=2*np.sqrt(w[1]),
angle=angle, alpha=0.7, color=next(prop_cycle), label="{0:.3f}".format(g[0]))
ax.add_patch(ell)
ax.autoscale()
ax.set_aspect('equal')
plt_configure(xlabel='x', ylabel='y')
plt.legend(loc='best')
plt.show()
