def learn_bg_model(
sequences,
num_mosaics=4,
order=3,
tolerance_per_base=7e-5
):
"""
@return: (bg_model, converted_sequences)
"""
bg_model = hmm.as_model(
create_mosaic_model(
num_mosaics=4,
p_transition=.1,
alphabet_size=4,
order=3,
dirichlet_prior_strength=.3
)
)
converted_seqs = [bg_model.converter.to_order_n(s) for s in sequences]
def _callback(LL):
logging.debug('Background model LL: %f', LL)
return True
tolerance=tolerance_per_base*sum(len(s) for s in sequences)
logging.info('Learning background model with %d mosaics of order %d, tolerance=%f', num_mosaics, order, tolerance)
start = time.time()
LL, iterations = bg_model.baum_welch(converted_seqs, tolerance=tolerance, callback=_callback)
logging.info('Achieved LL=%f after %d iterations and %f seconds', LL, iterations, time.time() - start)
return bg_model, converted_seqs
def uniform_bg_model():
bg_model = hmm.as_model(
create_mosaic_model(
num_mosaics=1,
p_transition=.1,
alphabet_size=4,
order=0,
dirichlet_prior_strength=None
)
)
return bg_model
