def run(n_documents=30, n_topics=5, n_vocabulary=10, n_words=50000, stochastic=False, maxiter=1000, seed=None):
if seed is not None:
np.random.seed(seed)
(corpus, word_documents) = generate_data(n_documents, n_topics, n_vocabulary, n_words)
if not stochastic:
Q = model(n_documents=n_documents, n_topics=n_topics, n_vocabulary=n_vocabulary,
corpus=corpus, word_documents=word_documents)
Q.update(repeat=maxiter)
else:
subset_size = 1000
Q = model(n_documents=n_documents, n_topics=n_topics, n_vocabulary=n_vocabulary,
corpus=corpus[:subset_size], word_documents=word_documents[:subset_size],
plates_multiplier=n_words/subset_size)
Q.ignore_bound_checks = True
delay = 1
forgetting_rate = 0.7
for n in range(maxiter):
# Observe a mini-batch
subset = np.random.choice(n_words, subset_size)
Q['words'].observe(corpus[subset])
Q['word_documents'].set_value(word_documents[subset])
# Learn intermediate variables
Q.update('topics')
# Set step length
step = (n + delay) ** (-forgetting_rate)
# Stochastic gradient for the global variables
Q.gradient_step('p_topic', 'p_word', scale=step)
bpplt.pyplot.figure()
bpplt.pyplot.plot(Q.L)
bpplt.pyplot.figure()
bpplt.hinton(Q['p_topic'])
bpplt.pyplot.title("Posterior topic distribution for each document")
bpplt.pyplot.xlabel("Topics")
bpplt.pyplot.ylabel("Documents")
bpplt.pyplot.figure()
bpplt.hinton(Q['p_word'])
bpplt.pyplot.title("Posterior word distributions for each topic")
bpplt.pyplot.xlabel("Words")
bpplt.pyplot.ylabel("Topics")
return
def _run(self, x, K=25, beta=0.5, alpha=0.00001, hinton_plot=False, end=False):
'''Only to be used when doing parameter optimization.'''
self.participant_list = x[0]
N = len(x[0]) #number of data points (i.e. WCS participants)
D = np.shape(x[1])[1] #number of features
#K = 20 #number of initial clusters
R = Dirichlet(K*[alpha],
name='R')
Z = Categorical(R,
plates=(N,1),
name='Z')
P = Beta([beta, beta],
plates=(D,K),
name='P')
X = Mixture(Z, Bernoulli, P)
Q = VB(Z, R, X, P)
P.initialize_from_random()
X.observe(x[1])
Q.update(repeat=1000)
log_likelihood = Q.L[Q.iter-1]
if hinton_plot:
bpplt.hinton(Z)
bpplt.pyplot.show()
bpplt.hinton(R)
bpplt.pyplot.show()
#Get the weight matrix stored in Z (weights determine which cluster data point belongs to)
z = Z._message_to_child()[0]
z = z * np.ones(Z.plates+(1,))
z = np.squeeze(z)
self.z = z
#Get the weights stored in R (proportional to the size of the clusters)
r = np.exp(R._message_to_child()[0])
r = r * np.ones(R.plates+(1,))
r = np.squeeze(r)
self.r = r
#Get the cluster assignment of each data point
self.c_assign = np.argmax(self.z, axis=1)
return log_likelihood
def generate_data(n_documents, n_topics, n_vocabulary, n_words):
# Generate random data from the generative model
# Generate document assignments for the words
word_documents = nodes.Categorical(np.ones(n_documents)/n_documents,
plates=(n_words,)).random()
# Topic distribution for each document
p_topic = nodes.Dirichlet(1e-1*np.ones(n_topics),
plates=(n_documents,)).random()
# Word distribution for each topic
p_word = nodes.Dirichlet(1e-1*np.ones(n_vocabulary),
plates=(n_topics,)).random()
# Topic for each word in each document
topic = nodes.Categorical(p_topic[word_documents],
plates=(n_words,)).random()
# Each word in each document
corpus = nodes.Categorical(p_word[topic],
plates=(n_words,)).random()
bpplt.pyplot.figure()
bpplt.hinton(p_topic)
bpplt.pyplot.title("True topic distribution for each document")
bpplt.pyplot.xlabel("Topics")
bpplt.pyplot.ylabel("Documents")
bpplt.pyplot.figure()
bpplt.hinton(p_word)
bpplt.pyplot.title("True word distributions for each topic")
bpplt.pyplot.xlabel("Words")
bpplt.pyplot.ylabel("Topics")
return (corpus, word_documents)
def generate_data(n_documents, n_topics, n_vocabulary, n_words):
# Generate random data from the generative model
# Generate document assignments for the words
word_documents = nodes.Categorical(np.ones(n_documents) / n_documents,
plates=(n_words, )).random()
# Topic distribution for each document
p_topic = nodes.Dirichlet(1e-1 * np.ones(n_topics),
plates=(n_documents, )).random()
# Word distribution for each topic
p_word = nodes.Dirichlet(1e-1 * np.ones(n_vocabulary),
plates=(n_topics, )).random()
# Topic for each word in each document
topic = nodes.Categorical(p_topic[word_documents],
plates=(n_words, )).random()
# Each word in each document
corpus = nodes.Categorical(p_word[topic], plates=(n_words, )).random()
bpplt.pyplot.figure()
bpplt.hinton(p_topic)
bpplt.pyplot.title("True topic distribution for each document")
bpplt.pyplot.xlabel("Topics")
bpplt.pyplot.ylabel("Documents")
bpplt.pyplot.figure()
bpplt.hinton(p_word)
bpplt.pyplot.title("True word distributions for each topic")
bpplt.pyplot.xlabel("Words")
bpplt.pyplot.ylabel("Topics")
return (corpus, word_documents)
import numpy as np
mu = np.array([[0, 0], [3, 4], [6, 0]])
std = 2.0
K = 3
N = 200
p0 = np.ones(K) / K
q = 0.9
r = (1 - q) / (K - 1)
P = q * np.identity(K) + r * (np.ones((3, 3)) - np.identity(3))
y = np.zeros((N, 2))
z = np.zeros(N)
state = np.random.choice(K, p=p0)
for n in range(N):
z[n] = state
y[n, :] = std * np.random.randn(2) + mu[state]
state = np.random.choice(K, p=P[state])
from bayespy.nodes import Dirichlet
a0 = Dirichlet(1e-3 * np.ones(K))
A = Dirichlet(1e-3 * np.ones((K, K)))
Z = CategoricalMarkovChain(a0, A, states=N)
Lambda = std**(-2) * np.identity(2)
from bayespy.nodes import Gaussian
Y = Mixture(Z, Gaussian, mu, Lambda)
Y.observe(y)
Q = VB(Y, Z, A, a0)
Q.update(repeat=1000)
bpplt.hinton(A)
bpplt.pyplot.show()
def run(self, K=25, beta=0.5, alpha=0.00001, foci_thresh=0, num_neigh=4, hinton_plot=False, end=False):
'''Performs one run of the BBDP according to the specified parameters.'''
print("Transforming WCS participant data into binary vectors...")
x = u.transform_data_all(self.langs, norm=False, end=end, foci=True, foci_thresh=foci_thresh, num_neigh=num_neigh)
print("Finished transforming participant data")
self.participant_list = x[0]
N = len(x[0]) #number of data points (i.e. WCS participants)
D = np.shape(x[1])[1] #number of features
#K = 20 #number of initial clusters
R = Dirichlet(K*[alpha],
name='R')
Z = Categorical(R,
plates=(N,1),
name='Z')
P = Beta([beta, beta],
plates=(D,K),
name='P')
X = Mixture(Z, Bernoulli, P)
Q = VB(Z, R, X, P)
P.initialize_from_random()
X.observe(x[1])
Q.update(repeat=1000)
if hinton_plot:
bpplt.hinton(Z)
bpplt.pyplot.show()
bpplt.hinton(R)
bpplt.pyplot.show()
#Get the weight matrix stored in Z (weights determine which cluster data point belongs to)
z = Z._message_to_child()[0]
z = z * np.ones(Z.plates+(1,))
z = np.squeeze(z)
self.z = z
#Get the weights stored in R (proportional to the size of the clusters)
r = np.exp(R._message_to_child()[0])
r = r * np.ones(R.plates+(1,))
r = np.squeeze(r)
self.r = r
#Get the cluster assignment of each data point
self.c_assign = np.argmax(self.z, axis=1)
#Write cluster results to a file
if self.write_to_file:
if end:
save_path = "cluster_results_end_K={}_B={}_a={}_t={}_nn={}".format(K, beta, alpha, foci_thresh, num_neigh)
else:
save_path = "cluster_results_K={}_B={}_a={}_t={}_nn={}".format(K, beta, alpha, foci_thresh, num_neigh)
while path.exists(save_path+".txt"):
#save_path already exists
try:
old_file_num = int(save_path[save_path.find('(')+1:-1])
new_file_num = old_file_num + 1
save_path = save_path[0:save_path.find('(')] + '(' + str(new_file_num) + ')'
except ValueError:
save_path = save_path + " (1)"
self.save_path = save_path
file = open(path.abspath(self.save_path+".txt"), 'w')
#Write cluster assignment matrix Z (gives the probability that observation i belongs to cluster j)
if 'Z' not in self.in_file:
for i in range(len(self.z)):
line = "\t".join([str(x) for x in self.z[i]]) + "\n"
file.write(line)
file.write('---Z\n')
self.in_file.append('Z')
#Write cluster weights matrix R (proportional to the size of the resulting clusters)
if 'R' not in self.in_file:
line = "\t".join([str(x) for x in self.r]) + "\n"
file.write(line)
file.write('---R\n')
self.in_file.append('R')
#Write deterministic cluster assignments with the corresponding participant key
if 'C' not in self.in_file:
line1 = "\t".join([str(x) for x in self.participant_list]) + "\n"
line2 = "\t".join([str(x) for x in self.c_assign]) + "\n"
file.write(line1)
file.write(line2)
file.write('---C\n')
self.in_file.append('C')
file.close()
return self.c_assign
def test_hinton_plot_dirichlet():
(R,P,Z) = _setup_bernoulli_mixture()
bpplt.hinton(R)
#( 5 : Parameter expansion 収束が遅い時) # from bayespy.inference.vmp import transformations # rotX = transformations.RotateGaussianARD(X) # rotC = transformations.RotateGaussianARD(C, alpha) # R = transformations.RotationOptimizer(rotC, rotX, D) # R.rotate() # alpha.initialize_from_prior() # C.initialize_from_prior() # X.initialize_from_parameters(np.random.randn(1, 100, D), 10) # tau.initialize_from_prior() # Q = VB(Y, C, X, alpha, tau) # Q.callback = R.rotate # Q.update(repeat=1000, tol=1e-6) # -----Examining the results----- # Plotting the results bpplt.pyplot.figure() bpplt.pdf(Q['tau'], np.linspace(60, 140, num=100)) V = Gaussian([3, 5], [[4, 2], [2, 5]]) bpplt.pyplot.figure() bpplt.contour(V, np.linspace(1, 5, num=100), np.linspace(3, 7, num=100)) bpplt.pyplot.figure() bpplt.hinton(C) bpplt.pyplot.figure() bpplt.plot(X, axis=-2) bpplt.pyplot.show()
import numpy numpy.random.seed(1) p0 = [0.1, 0.9, 0.1, 0.9, 0.1, 0.9, 0.1, 0.9, 0.1, 0.9] p1 = [0.1, 0.1, 0.1, 0.1, 0.1, 0.9, 0.9, 0.9, 0.9, 0.9] p2 = [0.9, 0.9, 0.9, 0.9, 0.9, 0.1, 0.1, 0.1, 0.1, 0.1] import numpy as np p = np.array([p0, p1, p2]) from bayespy.utils import random z = random.categorical([1 / 3, 1 / 3, 1 / 3], size=100) x = random.bernoulli(p[z]) N = 100 D = 10 K = 10 from bayespy.nodes import Categorical, Dirichlet R = Dirichlet(K * [1e-5], name='R') Z = Categorical(R, plates=(N, 1), name='Z') from bayespy.nodes import Beta P = Beta([0.5, 0.5], plates=(D, K), name='P') from bayespy.nodes import Mixture, Bernoulli X = Mixture(Z, Bernoulli, P) from bayespy.inference import VB Q = VB(Z, R, X, P) P.initialize_from_random() X.observe(x) Q.update(repeat=1000) import bayespy.plot as bpplt bpplt.hinton(R) bpplt.pyplot.show()
import numpy numpy.random.seed(1) p0 = [0.1, 0.9, 0.1, 0.9, 0.1, 0.9, 0.1, 0.9, 0.1, 0.9] p1 = [0.1, 0.1, 0.1, 0.1, 0.1, 0.9, 0.9, 0.9, 0.9, 0.9] p2 = [0.9, 0.9, 0.9, 0.9, 0.9, 0.1, 0.1, 0.1, 0.1, 0.1] import numpy as np p = np.array([p0, p1, p2]) from bayespy.utils import random z = random.categorical([1 / 3, 1 / 3, 1 / 3], size=100) x = random.bernoulli(p[z]) N = 100 D = 10 K = 10 from bayespy.nodes import Categorical, Dirichlet R = Dirichlet(K * [1e-5], name='R') Z = Categorical(R, plates=(N, 1), name='Z') from bayespy.nodes import Beta P = Beta([0.5, 0.5], plates=(D, K), name='P') from bayespy.nodes import Mixture, Bernoulli X = Mixture(Z, Bernoulli, P) from bayespy.inference import VB Q = VB(Z, R, X, P) P.initialize_from_random() X.observe(x) Q.update(repeat=1000) import bayespy.plot as bpplt bpplt.hinton(Z) bpplt.pyplot.show()
def test_hinton_plot_beta():
(R, P, Z) = _setup_bernoulli_mixture()
bpplt.hinton(P)
def test_hinton_plot_dirichlet():
(R, P, Z) = _setup_bernoulli_mixture()
bpplt.hinton(R)
def test_hinton_plot_categorical():
(R, P, Z) = _setup_bernoulli_mixture()
bpplt.hinton(Z)
N = 100 D = 10 K = 3 from bayespy.nodes import Categorical, Dirichlet R = Dirichlet(K * [1e-5], name='R') Z = Categorical(R, plates=(N, 1), name='Z') from bayespy.nodes import Beta P = Beta([0.5, 0.5], plates=(D, K), name='P') from bayespy.nodes import Mixture, Bernoulli X = Mixture(Z, Bernoulli, P) from bayespy.inference import VB Q = VB(Z, R, X, P) P.initialize_from_random() X.observe(x) Q.update(repeat=1000) import bayespy.plot as bpplt plt.figure() bpplt.hinton(R) plt.figure() bpplt.hinton(P) plt.figure() bpplt.hinton(p) plt.figure() bpplt.hinton(Z)
def test_hinton_plot_beta():
(R,P,Z) = _setup_bernoulli_mixture()
bpplt.hinton(P)
Q.update(repeat=1000)
#print(" P:")
#print( P.get_moments() )
#print(" R:")
#print( R.get_moments() )
print(" Z:")
print( Z.get_moments() )
print(" X:")
print( X.get_moments() )
bpplt.hinton(R)
#bpplt.hinton(P)
#bpplt.hinton(Z)
bpplt.pyplot.show()
#pp = pprint.PrettyPrinter(indent=4)
#pp.pprint(X)
def test_hinton_plot_categorical():
(R,P,Z) = _setup_bernoulli_mixture()
bpplt.hinton(Z)
import numpy numpy.random.seed(1) p0 = [0.1, 0.9, 0.1, 0.9, 0.1, 0.9, 0.1, 0.9, 0.1, 0.9] p1 = [0.1, 0.1, 0.1, 0.1, 0.1, 0.9, 0.9, 0.9, 0.9, 0.9] p2 = [0.9, 0.9, 0.9, 0.9, 0.9, 0.1, 0.1, 0.1, 0.1, 0.1] import numpy as np p = np.array([p0, p1, p2]) from bayespy.utils import random z = random.categorical([1 / 3, 1 / 3, 1 / 3], size=100) x = random.bernoulli(p[z]) N = 100 D = 10 K = 10 from bayespy.nodes import Categorical, Dirichlet R = Dirichlet(K * [1e-5], name='R') Z = Categorical(R, plates=(N, 1), name='Z') from bayespy.nodes import Beta P = Beta([0.5, 0.5], plates=(D, K), name='P') from bayespy.nodes import Mixture, Bernoulli X = Mixture(Z, Bernoulli, P) from bayespy.inference import VB Q = VB(Z, R, X, P) P.initialize_from_random() X.observe(x) Q.update(repeat=1000) import bayespy.plot as bpplt bpplt.hinton(P) bpplt.pyplot.show()
def run(n_documents=30,
n_topics=5,
n_vocabulary=10,
n_words=50000,
stochastic=False,
maxiter=1000,
seed=None):
if seed is not None:
np.random.seed(seed)
(corpus, word_documents) = generate_data(n_documents, n_topics,
n_vocabulary, n_words)
if not stochastic:
Q = model(n_documents=n_documents,
n_topics=n_topics,
n_vocabulary=n_vocabulary,
corpus=corpus,
word_documents=word_documents)
Q.update(repeat=maxiter)
else:
subset_size = 1000
Q = model(n_documents=n_documents,
n_topics=n_topics,
n_vocabulary=n_vocabulary,
corpus=corpus[:subset_size],
word_documents=word_documents[:subset_size],
plates_multiplier=n_words / subset_size)
Q.ignore_bound_checks = True
delay = 1
forgetting_rate = 0.7
for n in range(maxiter):
# Observe a mini-batch
subset = np.random.choice(n_words, subset_size)
Q['words'].observe(corpus[subset])
Q['word_documents'].set_value(word_documents[subset])
# Learn intermediate variables
Q.update('topics')
# Set step length
step = (n + delay)**(-forgetting_rate)
# Stochastic gradient for the global variables
Q.gradient_step('p_topic', 'p_word', scale=step)
bpplt.pyplot.figure()
bpplt.pyplot.plot(Q.L)
bpplt.pyplot.figure()
bpplt.hinton(Q['p_topic'])
bpplt.pyplot.title("Posterior topic distribution for each document")
bpplt.pyplot.xlabel("Topics")
bpplt.pyplot.ylabel("Documents")
bpplt.pyplot.figure()
bpplt.hinton(Q['p_word'])
bpplt.pyplot.title("Posterior word distributions for each topic")
bpplt.pyplot.xlabel("Words")
bpplt.pyplot.ylabel("Topics")
return
