def learn_params_mle(self, samples):
"""
Maximum liklihood estimation (MLE) parameter learing for a MRF.
Parameters
----------
samples: List
A list of fully observed samples for the spanning the total domain
of this MRF. Where samples[i][n] is the i'th sample for node n.
"""
samples = np.array(samples)
"""For every clique"""
for clq in self.cliques:
"""Obtain the evidence that is within the domain of this clique"""
local_samples = samples[:, clq.domain]
"""If there is evidence for this clique"""
if len(local_samples.tolist()) != 0:
"""Compute the counts of the samples"""
counts = general.compute_counts(local_samples, clq.pot.sizes)
"""Reshape the counts into a potentials lookup table"""
clq.unobserved_pot.T =\
general.mk_stochastic(np.array(counts, dtype=float))
clq.pot.T = clq.unobserved_pot.T.copy()
def learn_params_mle(self, samples):
"""
Maximum liklihood estimation (MLE) parameter learing for a MRF.
Parameters
----------
samples: List
A list of fully observed samples for the spanning the total domain
of this MRF. Where samples[i][n] is the i'th sample for node n.
"""
samples = np.array(samples)
"""For every clique"""
for clq in self.cliques:
"""Obtain the evidence that is within the domain of this clique"""
local_samples = samples[:, clq.domain]
"""If there is evidence for this clique"""
if len(local_samples.tolist()) != 0:
"""Compute the counts of the samples"""
counts = general.compute_counts(local_samples, clq.pot.sizes)
"""Reshape the counts into a potentials lookup table"""
clq.unobserved_pot.T =\
general.mk_stochastic(np.array(counts, dtype=float))
clq.pot.T = clq.unobserved_pot.T.copy()
def learn_params_mle(self, samples):
"""
Maximum liklihood estimation (MLE) parameter learing for a tabular
CPD.
Parameters
----------
samples: List
A list of fully observed samples for the spanning the total domain
of this CPD. Where samples[i][n] is the i'th sample for node n.
"""
"""Compute the counts of the samples"""
counts = general.compute_counts(samples, self.fam_size)
"""Reshape the counts into a CPT"""
self.CPT = general.mk_stochastic(np.array(counts, dtype=float))
def learn_params_mle(self, samples):
"""
Maximum liklihood estimation (MLE) parameter learing for a tabular
CPD.
Parameters
----------
samples: List
A list of fully observed samples for the spanning the total domain
of this CPD. Where samples[i][n] is the i'th sample for node n.
"""
"""Compute the counts of the samples"""
counts = general.compute_counts(samples, self.fam_size)
"""Reshape the counts into a CPT"""
self.CPT = general.mk_stochastic(np.array(counts, dtype=float))
def test_mk_stochastic():
"""
FUNCTION: mk_stochastic, in general.py
"""
"""Define input values"""
mat = np.random.random((5, 5, 5, 5, 5))
"""Execute function"""
mat = general.mk_stochastic(mat)
"""Assert that the sum over the last dimension is 1"""
ans = []
for i in range(0, 5):
for j in range(0, 5):
for k in range(0, 5):
for l in range(0, 5):
ans.append(np.sum(mat[i, j, k, l, :]))
assert not sum(np.round(np.array(ans), 3) != 1)
def test_mk_stochastic():
"""
FUNCTION: mk_stochastic, in general.py
"""
"""Define input values"""
mat = np.random.random((5, 5, 5, 5, 5))
"""Execute function"""
mat = general.mk_stochastic(mat)
"""Assert that the sum over the last dimension is 1"""
ans = []
for i in range(0, 5):
for j in range(0, 5):
for k in range(0, 5):
for l in range(0, 5):
ans.append(np.sum(mat[i, j, k, l, :]))
assert not sum(np.round(np.array(ans), 3) != 1)
def maximize_params(self):
"""
Maximize the parameters from the expected sufficent statistics.
"""
ess = np.array(self.ess).reshape(self.CPT.shape)
self.CPT = general.mk_stochastic(ess)
def maximize_params(self):
"""
Maximize the parameters from the expected sufficent statistics.
"""
ess = np.array(self.ess).reshape(self.unobserved_pot.T.shape)
self.unobserved_pot.T = general.mk_stochastic(ess)
