def subsample(self, fcs, *args, **kwargs):
x = fcs[:]
neg_py = mixnormpdf(x,
self.neg.pis,
self.neg.mus,
self.neg.sigmas,
logged=True,
*args,
**kwargs)
pos_py = mixnormpdf(x,
self.pos.pis,
self.pos.mus,
self.pos.sigmas,
logged=True,
*args,
**kwargs)
diff = pos_py - neg_py
probs = np.exp(diff)
probs = probs / np.sum(probs)
samp = npr.choice(np.arange(x.shape[0]),
size=self.n,
replace=False,
p=probs)
if isinstance(fcs, fcm.FCMdata):
node = SubsampleNode("", fcs.get_cur_node(), samp)
fcs.add_view(node)
return fcs
else:
return x[samp]
def testThatItWorks(self):
self.assertTrue(self._initialized)
## use kmeans to fit the data and get labels
kmeanResults = run_kmeans_with_sv(case1)
kmeanCentroids = kmeanResults['centroids']
kmeanLabels = kmeanResults['labels']
k = kmeanResults['k']
avgSilVal = kmeanResults['avgSilVal']
## ensure good kmean results
self.assertEqual(int(k),2)
self.assertEqual(len(kmeanCentroids),int(k))
self.assertEqual(len(kmeanLabels),int(case1.shape[0]))
## use sample stats to parameterize the mixnormpdf
mu = np.array([case1[np.where(kmeanLabels == i)[0],:].mean(axis=0) for i in [0,1]])
sigma = np.array([np.cov(case1[np.where(kmeanLabels == i)[0],:].T) for i in [0,1]])
pi = np.array([float(len(np.where(kmeanLabels == i)[0])) / float(len(kmeanLabels)) for i in [0,1]])
x = np.array([[8,20],[10,12]])
mnpdf = mixnormpdf(x,pi,mu,sigma)
## make sure the point the is far away has a smaller probability
self.assertTrue(mnpdf[1] > mnpdf[0])
## ensure we have proper probabilities
self.assertEqual(len(np.where(mnpdf < 1.0)[0]),2)
self.assertEqual(len(np.where(mnpdf > 0.0)[0]),2)
def subsample(self, fcs,*args, **kwargs):
x = fcs[:]
neg_py = mixnormpdf(x, self.neg.pis, self.neg.mus, self.neg.sigmas,
logged=True, *args, **kwargs)
pos_py = mixnormpdf(x, self.pos.pis, self.pos.mus, self.pos.sigmas,
logged=True,*args, **kwargs)
diff = pos_py - neg_py
probs = np.exp(diff)
probs = probs / np.sum(probs)
samp = npr.choice(np.arange(x.shape[0]), size=self.n,
replace=False, p=probs)
if isinstance(fcs, fcm.FCMdata):
node = SubsampleNode("", fcs.get_cur_node(), samp)
fcs.add_view(node)
return fcs
else:
return x[samp]
def subsample(self, fcs, *args, **kwargs):
x = fcs[:]
p = mixnormpdf(x, self.neg.pis, self.neg.mus, self.neg.sigmas)
p = 1/p
p = p/np.sum(p)
samp = npr.choice(np.arange(x.shape[0]), size=self.n, replace=False, p=p)
if isinstance(fcs, fcm.FCMdata):
node = SubsampleNode("", fcs.get_cur_node(), samp)
fcs.add_view(node)
return fcs
else:
return x[samp]
def subsample(self, fcs, *args, **kwargs):
x = fcs[:]
p = mixnormpdf(x, self.neg.pis, self.neg.mus, self.neg.sigmas)
p = 1 / p
p = p / np.sum(p)
samp = npr.choice(np.arange(x.shape[0]),
size=self.n,
replace=False,
p=p)
if isinstance(fcs, fcm.FCMdata):
node = SubsampleNode("", fcs.get_cur_node(), samp)
fcs.add_view(node)
return fcs
else:
return x[samp]
img = make_image(zip(x1, y1),view_xmin,view_xmax,view_ymin,view_ymax,
1024,1024)
xi = np.linspace(0,1024,1024)
yi = np.linspace(0,1024,1024)
plt.figure(figsize=(8,8))
# Pseudocolor using bilinear interpolation
plt.subplot(2,2,1)
try:
z = np.load('density_10072101_01.npy')
except:
m = stats.DPMixtureModel(nclusts=32, niter=100, burnin=1000)
m.ident = True
r = m.fit(data[:, cols], verbose=True).average()
z = stats.mixnormpdf(data[:, cols], r.pis(), r.mus(), r.sigmas())
np.save('density_10072101_02.npy', z)
idx = np.argsort(z)
plt.scatter(x[idx], y[idx], s=1, edgecolors='none', c=z[idx])
plt.xticks([])
plt.yticks([])
plt.imshow(z0, origin='lower', extent=[xmin, xmax, ymin, ymax],
cmap=plt.cm.binary)
plt.title('Pseudocolor (DPGMM)')
# Pseudocolor using kernel density estimate
plt.subplot(2,2,2)
kde = gaussian_kde([x1, y1])
z = kde.evaluate([x1, y1])
plt.scatter(x, y, s=1, edgecolors='none', c=z)
img = make_image(zip(x1, y1), view_xmin, view_xmax, view_ymin, view_ymax,
1024, 1024)
xi = np.linspace(0, 1024, 1024)
yi = np.linspace(0, 1024, 1024)
plt.figure(figsize=(8, 8))
# Pseudocolor using bilinear interpolation
plt.subplot(2, 2, 1)
try:
z = np.load('density_10072101_01.npy')
except:
m = stats.DPMixtureModel(nclusts=32, niter=100, burnin=1000)
m.ident = True
r = m.fit(data[:, cols], verbose=True).average()
z = stats.mixnormpdf(data[:, cols], r.pis(), r.mus(), r.sigmas())
np.save('density_10072101_02.npy', z)
idx = np.argsort(z)
plt.scatter(x[idx], y[idx], s=1, edgecolors='none', c=z[idx])
plt.xticks([])
plt.yticks([])
plt.imshow(z0,
origin='lower',
extent=[xmin, xmax, ymin, ymax],
cmap=plt.cm.binary)
plt.title('Pseudocolor (DPGMM)')
# Pseudocolor using kernel density estimate
plt.subplot(2, 2, 2)
kde = gaussian_kde([x1, y1])
