def initialize_pi(self, pi):
for i in range(1, self.K + 1):
for j in range(1, self.K + 1):
if i == j:
pi[i][j] = samplable.RealV(0.8)
else:
pi[i][j] = samplable.RealV(0.2/ (self.K - 1))
def initialize(self, dataset):
self.M = dataset.get_label_id_size()
print('The number of label ids = ' + str(self.M))
self.I = dataset.get_instance_size()
self.J = dataset.get_worker_size()
print('Total ' + str(self.I) + ' instances and ' + str(self.J) +
' workers')
self.initialK(dataset)
print('K = ' + str(self.K))
for i in range(1, self.I + 1):
inst = dataset.get_instance(i)
mcmldoc_inst = MCMLDOCInstance(inst)
mcmldoc_inst.initialize(self.M, self.K, self.R)
self.instances.append(mcmldoc_inst)
for w in range(1, self.J + 1):
worker = dataset.get_worker(w)
mcmldoc_worker = MCMLDOCWorker(worker)
mcmldoc_worker.initialize(self.M, self.K, self.R)
self.workers.append(mcmldoc_worker)
for r in range(1, self.R + 1):
theta_list = [None]
for m in range(1, self.M + 1):
thetas = numpy.ndarray(shape=(self.K + 1),
dtype=samplable.RealV,
order='C')
vals = utils.gen_rand_sum_one(self.K) # todo: can optimize
for i in range(1, self.K + 1):
thetas[i] = samplable.RealV(vals[i - 1])
theta_list.append(thetas)
self.theta_dict.setdefault(r, theta_list)
# initialize omega
if (len(self.omega) == 1):
vals = utils.gen_uniform_sum_one(self.R)
for r in range(1, self.R + 1):
self.omega.append(samplable.RealV(vals[r - 1]))
def initialize(self, M, K):
self.M = M
self.K = K
for m in range(1, self.M + 1):
self.y_list.append(samplable.IntV(0))
probs = numpy.ndarray(shape=(self.K + 1), dtype=samplable.RealV, order='C')
for i in range(1, self.K + 1):
probs[i] = samplable.RealV(1.0 / self.K)
self.y_prob_list.append(probs)
self.likelihood_list.append(samplable.RealV(0.0))
def initialize_pi_psi(self, pi, psi):
# initialize pi
for i in range(1, self.K + 1):
for j in range(1, self.K + 1):
## self.pi[i][j].setVal(1.0/self.K)
if i == j:
pi[i][j] = samplable.RealV(0.80)
else:
pi[i][j] = samplable.RealV(0.20 / (self.K - 1))
psi[i][j] = samplable.RealV(1.0)
def initialize(self, M, K, R):
self.M = M
self.K = K
self.R = R
for m in range(1, self.M + 1):
self.y_list.append(samplable.IntV(0))
self.y_combination.extend(utils.get_full_combination(self.M, self.K))
for i in range(1, len(self.y_combination)):
self.y_prob_list.append(samplable.RealV(0.0))
self.z = samplable.IntV(0)
for r in range(1, self.R + 1):
self.z_prob_list.append(samplable.RealV(0.0))
def random_initialize_pi(self, pi, diagonal_low, diagonal_high):
# we get K diagonal elements randomly in the range of low - high
diagonal = [None]
for i in range(1, self.K + 1):
val = random.uniform(diagonal_low, diagonal_high)
diagonal.append(val)
for i in range(1, self.K + 1):
remainder = 1.0 - diagonal[i]
parts = utils.split_val_rand(remainder, self.K - 1)
parts_index = 0
for j in range(1, self.K + 1):
if i == j:
pi[i][j] = samplable.RealV(diagonal[i])
else:
pi[i][j] = samplable.RealV(parts[parts_index])
parts_index += 1
def initialize(self, dataset):
self.M = dataset.get_label_id_size()
print('The number of label ids = ' + str(self.M))
self.I = dataset.get_instance_size()
self.J = dataset.get_worker_size()
print('Total ' + str(self.I) + ' instances and ' + str(self.J) +
' workers')
self.initialK(dataset)
print('K = ' + str(self.K))
for i in range(1, self.I + 1):
inst = dataset.get_instance(i)
mcoc_inst = MCOCInstance(inst)
mcoc_inst.initialize(self.M, self.K)
self.instances.append(mcoc_inst)
for w in range(1, self.J + 1):
worker = dataset.get_worker(w)
mcoc_worker = MCOCWorker(worker)
mcoc_worker.initialize(self.M, self.K)
self.workers.append(mcoc_worker)
for m in range(1, self.M + 1):
thetas = numpy.ndarray(shape=(self.K + 1),
dtype=samplable.RealV,
order='C')
for i in range(1, self.K + 1):
thetas[i] = samplable.RealV(1.0 / self.K)
self.theta_list.append(thetas)
def __init__(self, inst):
self.inst = inst
self.M = 0
self.K = 0
self.y_list = [None]
self.y_prob_list = [None]
self.likelihood = samplable.RealV(0.0)
self.y_combination = [None]
def initialize(self, M, K):
self.M = M
self.K = K
self.rho_list.append(None)
for m in range(1, self.M + 1):
rho = samplable.RealV(0.0)
self.initialize_rho(rho)
self.rho_list.append(rho)
def initialize(self, M, K, R):
self.M = M
self.K = K
self.R = R
# initialize rho dictionary
for r in range(1, self.R + 1):
rho_list = [None]
for m in range(1, self.M + 1):
rho = samplable.RealV(0.0)
self.random_initialize_rho(rho, 0.7, 0.9)
rho_list.append(rho)
self.rho_dict.setdefault(r, rho_list)
def initialize(self, M, K):
self.M = M
self.K = K
for m in range(1, self.M + 1):
self.y_list.append(samplable.IntV(0))
probs = numpy.ndarray(shape=(self.K + 1),
dtype=samplable.RealV,
order='C')
r = [numpy.random.random() for i in range(0, self.K)]
s = sum(r)
r = [i / s for i in r]
for i in range(1, self.K + 1):
probs[i] = samplable.RealV(r[i - 1])
self.y_prob_list.append(probs)
def initialize(self, dataset):
self.M = dataset.get_label_id_size()
print('The number of label ids = ' + str(self.M))
self.I = dataset.get_instance_size()
self.J = dataset.get_worker_size()
print('Total ' + str(self.I) + ' instances and ' + str(self.J) +
' workers')
self.initialK(dataset)
print('K = ' + str(self.K))
for m in range(1, self.M + 1):
lmd = numpy.ndarray(shape=(self.K + 1, self.K + 1),
dtype=samplable.RealV,
order='C')
prob = [None]
nu = [None]
for i in range(1, self.K + 1):
nu.append(samplable.RealV(1.0))
# \lmb diagonals bigger than the off-diagonals
for j in range(1, self.K + 1):
if i == j:
lmd[i][j] = samplable.RealV(1.5)
else:
lmd[i][j] = samplable.RealV(0.5)
prob.append(samplable.RealV(1.0 / self.K))
self.nu_list.append(nu)
self.lmd_list.append(lmd)
self.prob_list.append(prob)
for i in range(1, self.I + 1):
inst = dataset.get_instance(i)
ibcc_inst = IBCCInstance(inst)
ibcc_inst.initialize(self.M, self.K)
self.instances.append(ibcc_inst)
for w in range(1, self.J + 1):
worker = dataset.get_worker(w)
ibcc_worker = IBCCWorker(worker)
ibcc_worker.initialize(self.M, self.K)
self.workers.append(ibcc_worker)
def set_omega(self, vals):
for r in range(1, self.R + 1):
self.omega.append(samplable.RealV(vals[r]))
