def test_hand_made_data_set(self):
"Test HDPM with uncertainty on small hand-crafted data set."
numpy.random.seed(37)
F = 12
K = 5
genes = [
[
(0, [.1, .2, .99]),
],
[
(3, [.9, .9]),
(2, [.2, .99]),
(4, [.2, .99]),
(5, [.2, .99]),
],
[
(0, [.1, .2, .99]),
(2, [.1, .2]),
(10, [.1, .2, .99]),
(11, [.1, .2]),
(9, [.1, .2]),
],
]
options = U.get_default_options()
data = U.Data(genes, F, options)
dist = U.VariationalDistribution(data, K)
test_log_likelihood_per_update(dist)
stats = Statistics(dist)
stats.log()
def test_sampled_data(self):
"Test HDPM with uncertainty on sampled data sets of different sizes."
for i, (F, K, G, average_n_g) in enumerate((
( 2, 10, 20, 5),
( 2, 4, 20, 10),
( 2, 10, 20, 10),
( 2, 10, 220, 50),
(12, 80, 100, 50),
(80, 6, 200, 200),
)):
numpy.random.seed(i+1)
logging.debug('Testing sampled data with F=%d; K=%d; G=%d, average n_g=%d', F, K, G, average_n_g)
options = U.get_default_options()
options.a_tau = numpy.ones(F) / F
options.a_omega = numpy.ones(F) / F
rho = U.sample_rho(G, average_n_g=average_n_g)
sample = U.sample(options, rho, K, F)
genes = U.genes_from_sites(sample.sites, rho)
data = U.Data(genes, F, options)
dist = U.VariationalDistribution(data, K)
test_log_likelihood_per_update(dist)
reload(infpy.dp.hdpm.uncertainty)
import infpy.dp.hdpm.uncertainty as U
import infpy.dp.hdpm.uncertainty.summarise
reload(infpy.dp.hdpm.uncertainty.summarise)
from infpy.dp.hdpm.uncertainty.summarise import Statistics, InferenceHistory, Summariser
from infpy.convergence_test import LlConvergenceTest
logging.basicConfig(level=logging.DEBUG)
F, K, G, average_n_g = (12, 80, 100, 50)
numpy.random.seed(2)
logging.debug(
'Testing sampled data with F=%d; K=%d; G=%d, average n_g=%d', F, K, G, average_n_g)
options = U.get_default_options()
options.a_tau = numpy.ones(F)
options.a_omega = numpy.ones(F)
rho = U.sample_rho(G, average_n_g=average_n_g)
sample = U.sample(options, rho, K, F)
genes = U.genes_from_sites(sample.sites, rho)
data = U.Data(genes, F, options)
dist = U.VariationalDistribution(data, K)
Summariser(dist, 'output/sampled/summary').summarise_all()
history = InferenceHistory(dist)
LL = dist.log_likelihood()
LL_tolerance = 1e-8 * data.N
