K = 10 # Number of topics
V = 900 # Vocabulary Size
gamma = 0.25 # hyperparameter over doc-topic distribution
Defaults = dict()
Defaults['nDocTotal'] = 1000
Defaults['nWordsPerDoc'] = 500
# GLOBAL PROB DISTRIBUTION OVER TOPICS
trueBeta = np.ones(K)
trueBeta /= trueBeta.sum()
Defaults['topic_prior'] = gamma * trueBeta
# TOPIC by WORD distribution
Defaults['topics'] = Bars2D.Create2DBarsTopicWordParams(V,
K,
PRNG=PRNG,
fracMassOnTopic=0.999)
def get_data_info():
s = 'Clean Bars Data with %d true topics. Each doc uses 1-3 bars.' % (K)
return s
def get_data(seed=SEED, **kwargs):
''' Create toy dataset using bars topics.
Keyword Args
------------
seed : int
Determines pseudo-random generator used to make the toy data.
gamma = 0.85 # hyperparameter over doc-topic distribution
Defaults = dict()
Defaults['nDocTotal'] = 1000
Defaults['nWordsPerDoc'] = 300
# GLOBAL PROB DISTRIBUTION OVER TOPICS
B = 3.0
trueBeta = [B, 1, B, 1, B, 1, B, 1, B, 1.0]
trueBeta = np.asarray(trueBeta) / np.sum(trueBeta)
Defaults['topic_prior'] = gamma * trueBeta
# TOPIC by WORD distribution
Defaults['topics'] = Bars2D.Create2DBarsTopicWordParams2(
V,
K,
r=0.1,
PRNG=PRNG)
def get_data_info(**kwargs):
if 'nDocTotal' in kwargs:
nDocTotal = kwargs['nDocTotal']
else:
nDocTotal = Defaults['nDocTotal']
return 'Toy Bars2 Data. Ktrue=%d. nDocTotal=%d.' % (K, nDocTotal)
def get_data(**kwargs):
'''
Args