open journal (2/6/2017)

ldd

p = np.random.randint(0,10,[n_topics, n_words])*1.0
sum_over_topic = np.sum(p, axis=1)
dists = np.divide(p.T,sum_over_topic).T
n_topics = 3
n_words = 10
docs_topic = [0, 1, 2, 2, 1, 2, 0, 1, 2]
docs = []
for d_topic in docs_topic:
    _pvals = dists[d_topic]
#     print(np.sum(_pvals[:,-1]))
    docs.append(np.random.multinomial(n_words, _pvals, 1)[0])
from sklearn.decomposition import LatentDirichletAllocation as LDA
lda = LDA(3)
np.argmax(lda.fit_transform(docs),axis=1)
n_topics = 3
n_words = 10
docs_topic = [0, 1, 2, 2, 1, 2, 0, 1, 2]
sess = ed.get_session()
# DATA
pi_true = np.random.dirichlet(np.Vocab)
z_data = np.array([np.random.choice(K, 1, p=pi_true)[0] for n in range(N)])
print('pi={}'.format(pi_true))

# MODEL

topic_over_docs = Dirichlet(tf.ones(n_words))
w_ = Dirichlet(tf.ones(n_words))
w_ = Categorical(probs= pi, sample_shape=N)

# INFERENCE
qpi = Dirichlet(tf.nn.softmax(tf.Variable(tf.random_normal([K]))))

inference = ed.KLqp({pi: qpi}, data={z: z_data})
inference.run(n_iter=15000, n_samples=30)

print('Inferred pi={}'.format(sess.run(qpi.mean())))