def test_Fisher(self):
"""Test Fisher scoring estimation for hierarchical regression"""
#mle_sequential = glm.ordinal_estimation(ratio='sequential', distribution = core.NormalDistribution(), data = self._data_ordinal, response = 0, explanatories = {1,2,3,4})
mle = glm.ordinal_estimation(ratio='sequential', data = self._data_ordinal, response = 0, explanatories = {1,2,3,4})
a = mle.estimated.predictor.alpha
mle = glm.hierarchical_estimation(data = self._data_hierarchical, response = 0, explanatories = {1,2,3,4})
A0 = mle[""].estimated.predictor.alpha[0]
A1 = mle["12"].estimated.predictor.alpha[0]
self.assertAlmostEqual(a[0], A0, places=3)
self.assertAlmostEqual(a[1], A1, places=2)
mle_sequential_cauchy = glm.ordinal_estimation(ratio='sequential', distribution = core.CauchyDistribution(), data = self._data_ordinal, response = 0, explanatories = {1,2,3,4})
estimator = glm.nominal_estimation(distribution = core.CauchyDistribution())
mle_hierarchical = glm.hierarchical_estimation(default = estimator, data = self._data_hierarchical, response = 0, explanatories = {1,2,3,4})
alpha_sequential_cauchy = mle_sequential_cauchy.estimated.predictor.alpha
alpha_hierarchical_root = mle_hierarchical[""].estimated.predictor.alpha[0]
alpha_hierarchical_12 = mle_hierarchical["12"].estimated.predictor.alpha[0]
self.assertAlmostEqual(alpha_sequential_cauchy[0], alpha_hierarchical_root, places=2)
self.assertAlmostEqual(alpha_sequential_cauchy[1], alpha_hierarchical_12, places=2)
mle_sequential_laplace = glm.ordinal_estimation(ratio='sequential', distribution = core.LaplaceDistribution(), data = self._data_ordinal, response = 0, explanatories = {1,2,3,4})
estimator_12 = glm.nominal_estimation(distribution = core.LaplaceDistribution())
mle_hierarchical = glm.hierarchical_estimation(default = estimator, **{"12" : estimator_12}, data = self._data_hierarchical, response = 0, explanatories = {1,2,3,4})
alpha_sequential_laplace = mle_sequential_laplace.estimated.predictor.alpha
alpha_hierarchical_root = mle_hierarchical[""].estimated.predictor.alpha[0]
alpha_hierarchical_12 = mle_hierarchical["12"].estimated.predictor.alpha[0]
self.assertAlmostEqual(alpha_sequential_cauchy[0], alpha_hierarchical_root, places=2)
self.assertAlmostEqual(alpha_sequential_laplace[1], alpha_hierarchical_12, places=1)
binary_estimator_cauchy = glm.binary_estimation(distribution = core.CauchyDistribution())
binary_estimator_laplace = glm.binary_estimation(distribution = core.LaplaceDistribution())
mle_hierarchical = glm.hierarchical_estimation(default = binary_estimator_cauchy, **{"12" : binary_estimator_laplace}, data = self._data_hierarchical, response = 0, explanatories = {1,2,3,4})
alpha_hierarchical_root = mle_hierarchical[""].estimated.predictor.alpha
alpha_hierarchical_12 = mle_hierarchical["12"].estimated.predictor.alpha
self.assertAlmostEqual(alpha_sequential_cauchy[0], alpha_hierarchical_root, places=3)
def test_Fisher(self):
"""Test Fisher scoring estimation for canonical ordinal regression with complete design"""
mle = glm.ordinal_estimation(data=self._data, response = 0, explanatories = {1, 2})
alpha = mle.estimated.predictor.alpha
delta = mle.estimated.predictor.delta
self.assertAlmostEqual(mle.loglikelihood, self._loglikelihood, places=2)
self._data.AirConditioning.sample_space = self._data.AirConditioning.sample_space.as_nominal()
mle = glm.nominal_estimation(data=self._data, response = 0, explanatories = {1, 2})
self.assertAlmostEqual(mle.loglikelihood, self._loglikelihood, places=2)
def test_Fisher(self):
"""Test Fisher scoring estimation for ordinal regression with proportional design"""
mle = glm.ordinal_estimation(ratio='sequential', Z='constrained', data=self._data, response = 0, explanatories = {1,2,3,4}, partial_proportional={1,2,3})
alpha = mle.estimated.predictor.alpha
delta = mle.estimated.predictor.delta
for i in range(len(self._alpha)):
x = int(alpha[i]*1000)/float(1000)
self.assertEqual(x, self._alpha[i])
for i in range(len(self._delta)):
x = int(delta[i]*1000)/float(1000)
self.assertEqual(x, self._delta[i])
def test_Fisher(self):
"""Test Fisher scoring estimation for cumulative regression with proportional design"""
#import ipdb
#ipdb.set_trace()
# beta_init = linalg.Vector([-0.66, 0.66, 0., 0., 0.])
# mle = glm.ordinal_estimation(ratio='cumulative', Z='proportional', data=self._data, response = 0, explanatories = {1,2}, beta_init=beta_init)
mle = glm.ordinal_estimation(ratio='cumulative', Z='proportional', data=self._data, response = 0, explanatories = {1, 2})
alpha = mle.estimated.predictor.alpha
delta = mle.estimated.predictor.delta
self.assertAlmostEqual(mle.loglikelihood, self._loglikelihood, places=3)
for i in range(len(self._alpha)):
self.assertAlmostEqual(alpha[i], self._alpha[i], places=3)
for i in range(len(self._delta)):
self.assertAlmostEqual(delta[i], self._delta[i], places=3)
def test_Fisher(self):
"""Test Fisher scoring estimation for ordinal regression with proportional design"""
mle = glm.ordinal_estimation(ratio='sequential',
Z='constrained',
data=self._data,
response=0,
explanatories={1, 2, 3, 4},
partial_proportional={1, 2, 3})
alpha = mle.estimated.predictor.alpha
delta = mle.estimated.predictor.delta
for i in range(len(self._alpha)):
x = int(alpha[i] * 1000) / float(1000)
self.assertEqual(x, self._alpha[i])
for i in range(len(self._delta)):
x = int(delta[i] * 1000) / float(1000)
self.assertEqual(x, self._delta[i])
def test_Fisher(self):
"""Test Fisher scoring estimation for canonical ordinal regression with complete design"""
mle = glm.ordinal_estimation(data=self._data,
response=0,
explanatories={1, 2})
alpha = mle.estimated.predictor.alpha
delta = mle.estimated.predictor.delta
self.assertAlmostEqual(mle.loglikelihood,
self._loglikelihood,
places=2)
self._data.AirConditioning.sample_space = self._data.AirConditioning.sample_space.as_nominal(
)
mle = glm.nominal_estimation(data=self._data,
response=0,
explanatories={1, 2})
self.assertAlmostEqual(mle.loglikelihood,
self._loglikelihood,
places=2)
def test_Fisher(self):
"""Test Fisher scoring estimation for cumulative regression with proportional design"""
#import ipdb
#ipdb.set_trace()
# beta_init = linalg.Vector([-0.66, 0.66, 0., 0., 0.])
# mle = glm.ordinal_estimation(ratio='cumulative', Z='proportional', data=self._data, response = 0, explanatories = {1,2}, beta_init=beta_init)
mle = glm.ordinal_estimation(ratio='cumulative',
Z='proportional',
data=self._data,
response=0,
explanatories={1, 2})
alpha = mle.estimated.predictor.alpha
delta = mle.estimated.predictor.delta
self.assertAlmostEqual(mle.loglikelihood,
self._loglikelihood,
places=3)
for i in range(len(self._alpha)):
self.assertAlmostEqual(alpha[i], self._alpha[i], places=3)
for i in range(len(self._delta)):
self.assertAlmostEqual(delta[i], self._delta[i], places=3)
def test_Fisher(self):
"""Test Fisher scoring estimation for hierarchical regression"""
#mle_sequential = glm.ordinal_estimation(ratio='sequential', distribution = core.NormalDistribution(), data = self._data_ordinal, response = 0, explanatories = {1,2,3,4})
mle = glm.ordinal_estimation(ratio='sequential',
data=self._data_ordinal,
response=0,
explanatories={1, 2, 3, 4})
a = mle.estimated.predictor.alpha
mle = glm.hierarchical_estimation(data=self._data_hierarchical,
response=0,
explanatories={1, 2, 3, 4})
A0 = mle[""].estimated.predictor.alpha[0]
A1 = mle["12"].estimated.predictor.alpha[0]
self.assertAlmostEqual(a[0], A0, places=3)
self.assertAlmostEqual(a[1], A1, places=2)
mle_sequential_cauchy = glm.ordinal_estimation(
ratio='sequential',
distribution=core.CauchyDistribution(),
data=self._data_ordinal,
response=0,
explanatories={1, 2, 3, 4})
estimator = glm.nominal_estimation(
distribution=core.CauchyDistribution())
mle_hierarchical = glm.hierarchical_estimation(
default=estimator,
data=self._data_hierarchical,
response=0,
explanatories={1, 2, 3, 4})
alpha_sequential_cauchy = mle_sequential_cauchy.estimated.predictor.alpha
alpha_hierarchical_root = mle_hierarchical[
""].estimated.predictor.alpha[0]
alpha_hierarchical_12 = mle_hierarchical[
"12"].estimated.predictor.alpha[0]
self.assertAlmostEqual(alpha_sequential_cauchy[0],
alpha_hierarchical_root,
places=2)
self.assertAlmostEqual(alpha_sequential_cauchy[1],
alpha_hierarchical_12,
places=2)
mle_sequential_laplace = glm.ordinal_estimation(
ratio='sequential',
distribution=core.LaplaceDistribution(),
data=self._data_ordinal,
response=0,
explanatories={1, 2, 3, 4})
estimator_12 = glm.nominal_estimation(
distribution=core.LaplaceDistribution())
mle_hierarchical = glm.hierarchical_estimation(
default=estimator,
**{"12": estimator_12},
data=self._data_hierarchical,
response=0,
explanatories={1, 2, 3, 4})
alpha_sequential_laplace = mle_sequential_laplace.estimated.predictor.alpha
alpha_hierarchical_root = mle_hierarchical[
""].estimated.predictor.alpha[0]
alpha_hierarchical_12 = mle_hierarchical[
"12"].estimated.predictor.alpha[0]
self.assertAlmostEqual(alpha_sequential_cauchy[0],
alpha_hierarchical_root,
places=2)
self.assertAlmostEqual(alpha_sequential_laplace[1],
alpha_hierarchical_12,
places=1)
binary_estimator_cauchy = glm.binary_estimation(
distribution=core.CauchyDistribution())
binary_estimator_laplace = glm.binary_estimation(
distribution=core.LaplaceDistribution())
mle_hierarchical = glm.hierarchical_estimation(
default=binary_estimator_cauchy,
**{"12": binary_estimator_laplace},
data=self._data_hierarchical,
response=0,
explanatories={1, 2, 3, 4})
alpha_hierarchical_root = mle_hierarchical[
""].estimated.predictor.alpha
alpha_hierarchical_12 = mle_hierarchical[
"12"].estimated.predictor.alpha
self.assertAlmostEqual(alpha_sequential_cauchy[0],
alpha_hierarchical_root,
places=3)