def Py_z2(data, i, clustering, cluster_assn, prior, ti=0):
Y = data[1, 0]
T = data[1, 1]
if type(i) == int:
new_y = Y[i]
new_t = T[:, i]
else:
new_y = i
new_t = ti
## NOTE: Should change var1, l and var2
v1 = prior["var1"] # 0.1
v2 = prior["var2"] # 0.2
l = prior["length_scale"] # 0.05
GPs = [
GPCluster(rbf_kern,
Y=Y[list(cluster)],
T=T[:, list(cluster)],
var1=v1,
l=l,
var2=v2) for cluster in clustering
]
likelihoods = [GPs[j].likelihood(new_y, new_t) for j in range(len(GPs))]
empty_gp = GPCluster(rbf_kern,
Y=np.array([0]),
T=np.array([0]),
var1=v1,
l=l,
var2=v2)
likelihoods.append(empty_gp.likelihood_empty(new_y, new_t))
return np.array(likelihoods)
dp = DCRP(alpha1, alpha2, Px_z1, Py_z2, prior1, prior2)
dp._initialize_assn(train_data)
print "Number of clusters for X = ", len(dp.c1)
print "Number of clusters for Y = ", len(dp.c2)
# Run gibbs sampling
dp._gibbs_sampling_crp(train_data)
print "Number of clusters for X = ", len(dp.c1)
print "Number of clusters for Y = ", len(dp.c2)
# Final clustering
dp_c2 = len(dp.c2)
dp_gp = []
for k in range(dp_c2):
dp_gp.append(GPCluster(rbf_kern,Y=train_data[1,0][list(dp.c2[k]),:],\
T=train_data[1,1][:,list(dp.c2[k])],var1=prior2["var1"], \
l=prior2["length_scale"],var2=prior2["var2"]))
dp_gp.append(GPCluster(rbf_kern,Y=np.array([[0,0,0,0,0]]),\
T=np.array([[0,1,2,3,4]]).T,var1=prior2["var1"],\
l=prior2["length_scale"],var2=prior2["var2"]))
ms = 0
for m in range(size_test):
y = test_data[1, 0][m].reshape((-1, 1))
y = y[:4]
t = np.array([[0, 1, 2, 3]])
x = test_data[0, 0][m]
prob = dp.predict(x, y, t, train_data)
t = np.array([[0, 1, 2, 3, 4]])
sum = 0
for i in range(prob.shape[0]):