def test_merging(self, num_mixands=100, max_num_mixands=10, spread=4,
speak=False):
if max_num_mixands is None:
animate = True
max_num_mixands = num_mixands
else:
animate = False
# Generate the unmerged and merged gaussians
weights, means, covariances = generate_random_params(num_mixands,
ndims=2,
spread=spread)
unmerged_gauss_2d = GaussianMixture(weights.copy(),
means.copy(),
covariances.copy(),
max_num_mixands=len(weights))
merged_gauss_2d = GaussianMixture(weights.copy(),
means.copy(),
covariances.copy(),
max_num_mixands=max_num_mixands)
matlab_merged_gauss_2d = self.matlab_gm(weights.copy(),
means.copy(),
covariances.copy(),
max_num_mixands)
mixtures = {'unmerged': unmerged_gauss_2d,
'merged': merged_gauss_2d,
'matlab merged': matlab_merged_gauss_2d
}
# Setup figure and levels
fig = plt.figure(figsize=(18,6))
axes = []
_, max_1 = unmerged_gauss_2d.max_point_by_grid()
_, max_2 = merged_gauss_2d.max_point_by_grid()
_, max_3 = matlab_merged_gauss_2d.max_point_by_grid()
max_prob = np.max((max_1, max_2, max_3))
levels = np.linspace(0, max_prob * 1.2, 50)
# Plot all three
ax = fig.add_subplot(131)
axes.append(ax)
title = 'Original GM ({} mixands)'\
.format(unmerged_gauss_2d.weights.size)
unmerged_gauss_2d.plot(ax=ax, levels=levels, title=title)
ax = fig.add_subplot(132)
axes.append(ax)
title = 'Python Merged GM ({} mixands)'\
.format(merged_gauss_2d.weights.size)
merged_gauss_2d.plot(ax=ax, levels=levels, title=title)
ax = fig.add_subplot(133)
axes.append(ax)
title = 'Matlab Merged GM ({} mixands)'\
.format(matlab_merged_gauss_2d.weights.size)
matlab_merged_gauss_2d.plot(ax=ax, levels=levels, title=title)
# Add a colorbar
fig.subplots_adjust(right=0.85)
cbar_ax = fig.add_axes([0.875, 0.1, 0.025, 0.8])
fig.colorbar(unmerged_gauss_2d.contourf, cax=cbar_ax)
class GMAnimation(object):
"""docstring for merged_gm"""
def __init__(self, parent, mixand_rate=2, levels=None, axes=None):
self.max_num_mixands = mixtures['unmerged'].weights.size
self.num_mixands = 1
self.mixand_rate = mixand_rate
self.levels = levels
self.axes = axes
self.parent = parent
def update(self,i=0):
# Regenerate GMs
merged_gauss_2d = GaussianMixture(weights, means, covariances,
max_num_mixands=self.num_mixands)
matlab_merged_gauss_2d = self.parent.matlab_gm(weights,
means,
covariances,
max_num_mixands=self.num_mixands)
# Replot GMs
title = 'Python Merged GM ({} mixands)'.format(merged_gauss_2d.weights.size)
if hasattr(self,'old_contour'):
merged_gauss_2d.contourf = self.old_contour
merged_gauss_2d.plot_remove()
self.old_contour = merged_gauss_2d.plot(ax=self.axes[1], levels=self.levels, title=title)
title = 'Matlab Merged GM ({} mixands)'.format(matlab_merged_gauss_2d.weights.size)
if hasattr(self,'old_matlab_contour'):
matlab_merged_gauss_2d.contourf = self.old_matlab_contour
matlab_merged_gauss_2d.plot_remove()
self.old_matlab_contour = matlab_merged_gauss_2d.plot(ax=self.axes[2], levels=self.levels, title=title)
# Decrement mixands (with wrapping)
if self.num_mixands == self.max_num_mixands:
self.num_mixands = 1
elif np.int(self.num_mixands * self.mixand_rate) < self.max_num_mixands:
self.num_mixands = np.int(self.num_mixands * self.mixand_rate)
else:
self.num_mixands = self.max_num_mixands
def compare_results():
pass
if animate:
gm_ani = GMAnimation(self, mixand_rate=2, levels=levels, axes=axes)
ani = animation.FuncAnimation(fig, gm_ani.update,
interval=100,
repeat=True,
blit=False,
)
else:
self.diff(merged_gauss_2d, matlab_merged_gauss_2d)
plt.show()
self.check_diff()
def test_merging(self,
num_mixands=100,
max_num_mixands=10,
spread=4,
speak=False):
if max_num_mixands is None:
animate = True
max_num_mixands = num_mixands
else:
animate = False
# Generate the unmerged and merged gaussians
weights, means, covariances = generate_random_params(num_mixands,
ndims=2,
spread=spread)
unmerged_gauss_2d = GaussianMixture(weights.copy(),
means.copy(),
covariances.copy(),
max_num_mixands=len(weights))
merged_gauss_2d = GaussianMixture(weights.copy(),
means.copy(),
covariances.copy(),
max_num_mixands=max_num_mixands)
matlab_merged_gauss_2d = self.matlab_gm(weights.copy(), means.copy(),
covariances.copy(),
max_num_mixands)
mixtures = {
'unmerged': unmerged_gauss_2d,
'merged': merged_gauss_2d,
'matlab merged': matlab_merged_gauss_2d
}
# Setup figure and levels
fig = plt.figure(figsize=(18, 6))
axes = []
_, max_1 = unmerged_gauss_2d.max_point_by_grid()
_, max_2 = merged_gauss_2d.max_point_by_grid()
_, max_3 = matlab_merged_gauss_2d.max_point_by_grid()
max_prob = np.max((max_1, max_2, max_3))
levels = np.linspace(0, max_prob * 1.2, 50)
# Plot all three
ax = fig.add_subplot(131)
axes.append(ax)
title = 'Original GM ({} mixands)'\
.format(unmerged_gauss_2d.weights.size)
unmerged_gauss_2d.plot(ax=ax, levels=levels, title=title)
ax = fig.add_subplot(132)
axes.append(ax)
title = 'Python Merged GM ({} mixands)'\
.format(merged_gauss_2d.weights.size)
merged_gauss_2d.plot(ax=ax, levels=levels, title=title)
ax = fig.add_subplot(133)
axes.append(ax)
title = 'Matlab Merged GM ({} mixands)'\
.format(matlab_merged_gauss_2d.weights.size)
matlab_merged_gauss_2d.plot(ax=ax, levels=levels, title=title)
# Add a colorbar
fig.subplots_adjust(right=0.85)
cbar_ax = fig.add_axes([0.875, 0.1, 0.025, 0.8])
fig.colorbar(unmerged_gauss_2d.contourf, cax=cbar_ax)
class GMAnimation(object):
"""docstring for merged_gm"""
def __init__(self, parent, mixand_rate=2, levels=None, axes=None):
self.max_num_mixands = mixtures['unmerged'].weights.size
self.num_mixands = 1
self.mixand_rate = mixand_rate
self.levels = levels
self.axes = axes
self.parent = parent
def update(self, i=0):
# Regenerate GMs
merged_gauss_2d = GaussianMixture(
weights,
means,
covariances,
max_num_mixands=self.num_mixands)
matlab_merged_gauss_2d = self.parent.matlab_gm(
weights,
means,
covariances,
max_num_mixands=self.num_mixands)
# Replot GMs
title = 'Python Merged GM ({} mixands)'.format(
merged_gauss_2d.weights.size)
if hasattr(self, 'old_contour'):
merged_gauss_2d.contourf = self.old_contour
merged_gauss_2d.plot_remove()
self.old_contour = merged_gauss_2d.plot(ax=self.axes[1],
levels=self.levels,
title=title)
title = 'Matlab Merged GM ({} mixands)'.format(
matlab_merged_gauss_2d.weights.size)
if hasattr(self, 'old_matlab_contour'):
matlab_merged_gauss_2d.contourf = self.old_matlab_contour
matlab_merged_gauss_2d.plot_remove()
self.old_matlab_contour = matlab_merged_gauss_2d.plot(
ax=self.axes[2], levels=self.levels, title=title)
# Decrement mixands (with wrapping)
if self.num_mixands == self.max_num_mixands:
self.num_mixands = 1
elif np.int(self.num_mixands *
self.mixand_rate) < self.max_num_mixands:
self.num_mixands = np.int(self.num_mixands *
self.mixand_rate)
else:
self.num_mixands = self.max_num_mixands
def compare_results():
pass
if animate:
gm_ani = GMAnimation(self, mixand_rate=2, levels=levels, axes=axes)
ani = animation.FuncAnimation(
fig,
gm_ani.update,
interval=100,
repeat=True,
blit=False,
)
else:
self.diff(merged_gauss_2d, matlab_merged_gauss_2d)
plt.show()
self.check_diff()
