return np.sum(topiclib.partial_slda_local_elbo(v.labeled[d], v.y[d], v.alphaL, v.beta[-v.Kl:], v.gammaL[d], v.phiL[d], v.eta, v.sigma_squared) for d in xrange(len(v.labeled)))
run_tlc = partial(graphlib.run_variational_em,
e_step_func=tlc_e_step,
m_step_func=tlc_m_step,
global_elbo_func=tlc_global_elbo,
print_func=tlc_print_func)
if __name__=='__main__':
dirname = 'synthtlc'
dirname = 'synthbig'
# use my tlc synthetically generated dataset
documents = topiclib.read_sparse(dirname + '/documents.dat')
comments = topiclib.read_sparse(dirname + '/comments.dat')
labeled_documents = topiclib.read_sparse(dirname + '/labeled.dat')
background = topiclib.read_sparse(dirname + '/background.dat')
y = np.loadtxt(dirname + '/yL.npy')
real_data = (documents, comments, labeled_documents, background, y)
var = TLCVars(real_data, Ku=29, Ks=5, Kb=24)
try:
output = run_tlc(var)
except Exception,e:
print e
import pdb; pdb.post_mortem()
[
1.7,
2.0,
1.2,
4.8,
5,
4.2,
])
#var = SupervisedLDAVars(test_data, K=3)
#var = SupervisedLDAVars(noisy_test_data, K=3)
# use my big generated dataset
n = 9994
labeled_documents = topiclib.read_sparse('data/synthbigtlc/labeled.dat')[:100]
y = np.loadtxt('data/synthbigtlc/yL.npy')[:100]
real_data = (labeled_documents, y)
var = PartialSupervisedLDAVars(real_data, Ks=5, Kb=20)
try:
output = run_partial_slda(var)
except Exception,e:
print e
import pdb; pdb.post_mortem()
[(5,2), (6,1), (8,1), (9,1),],
],
[
1.7,
2.0,
1.2,
4.8,
5,
4.2,
])
#var = SupervisedLDAVars(test_data, K=3)
#var = SupervisedLDAVars(noisy_test_data, K=3)
# use my big generated dataset
labeled_documents = topiclib.read_sparse('synthtlc/labeled.dat')
y = np.loadtxt('synthtlc/yL.npy')
real_data = (labeled_documents, y)
var = SupervisedLDAVars(real_data, K=13)
try:
output = run_slda(var)
except Exception,e:
print e
import pdb; pdb.post_mortem()