def test_fit_prior_sympy(self):
# carry out fit
S = bl.Study()
S.loadData(np.array([1, 2, 3, 4, 5]))
S.setOM(
bl.om.Poisson('rate',
bl.oint(0, 6, 1000),
prior=stats.Exponential('expon', 1.)))
S.setTM(bl.tm.GaussianRandomWalk('sigma', 0.1, target='rate'))
S.fit()
# test parameter distributions
np.testing.assert_allclose(
S.getParameterDistributions('rate', density=False)[1][:, 250],
[0.0020607, 0.0019692, 0.00187339, 0.0018053, 0.00179584],
rtol=1e-05,
err_msg='Erroneous posterior distribution values.')
# test parameter mean values
np.testing.assert_allclose(
S.getParameterMeanValues('rate'),
[2.25427356, 2.26949283, 2.28527551, 2.29704214, 2.30024139],
rtol=1e-05,
err_msg='Erroneous posterior mean values.')
# test model evidence value
np.testing.assert_almost_equal(S.logEvidence,
-11.1819034242,
decimal=5,
err_msg='Erroneous log-evidence value.')
def test_fit_prior_sympy(self):
# carry out fit
S = bl.Study()
S.loadData(np.array([1, 2, 3, 4, 5]))
S.setOM(
bl.om.Gaussian(
'mean',
bl.cint(0, 6, 20),
'sigma',
bl.oint(0, 2, 20),
prior=[stats.Uniform('u', 0, 6),
stats.Exponential('e', 2.)]))
S.setTM(bl.tm.GaussianRandomWalk('sigma', 0.1, target='mean'))
S.fit()
# test parameter distributions
np.testing.assert_allclose(
S.getParameterDistributions('mean', density=False)[1][:, 5],
[0.00909976, 0.0089861, 0.00887967, 0.00881235, 0.00880499],
rtol=1e-05,
err_msg='Erroneous posterior distribution values.')
# test parameter mean values
np.testing.assert_allclose(
S.getParameterMeanValues('mean'),
[2.9942575, 2.99646768, 3., 3.00353232, 3.0057425],
rtol=1e-05,
err_msg='Erroneous posterior mean values.')
# test model evidence value
np.testing.assert_almost_equal(S.logEvidence,
-12.4324853153,
decimal=5,
err_msg='Erroneous log-evidence value.')
def test_optimize(self):
# carry out fit
S = bl.Study()
S.loadData(np.array([1, 2, 3, 4, 5]))
S.setOM(bl.om.Poisson('rate', bl.oint(0, 6, 1000), prior=stats.Exponential('expon', 1.)))
T = bl.tm.CombinedTransitionModel(bl.tm.GaussianRandomWalk('sigma', 2.1, target='rate'),
bl.tm.RegimeSwitch('log10pMin', -3))
S.setTM(T)
S.optimize()
# test parameter distributions
np.testing.assert_allclose(S.getParameterDistributions('rate', density=False)[1][:, 250],
[0.00181567, 0.00213315, 0.00091028, 0.00041154, 0.00090885],
rtol=1e-05, err_msg='Erroneous posterior distribution values.')
# test parameter mean values
np.testing.assert_allclose(S.getParameterMeanValues('rate'),
[1.01204314, 2.25763551, 3.24176817, 3.74634864, 3.12632199],
rtol=1e-05, err_msg='Erroneous posterior mean values.')
# test model evidence value
np.testing.assert_almost_equal(S.logEvidence, -9.47362827569, decimal=5,
err_msg='Erroneous log-evidence value.')
# test optimized hyper-parameter values
np.testing.assert_almost_equal(S.getHyperParameterValue('sigma'), 2.11216289063, decimal=5,
err_msg='Erroneous log-evidence value.')
np.testing.assert_almost_equal(S.getHyperParameterValue('log10pMin'), -3.0, decimal=5,
err_msg='Erroneous log-evidence value.')
def test_step_add2TM_2hp_prior_hyperpriors_TMprior(self):
# carry out fit
S = bl.OnlineStudy(storeHistory=True)
S.setOM(bl.om.Gaussian('mean', bl.cint(0, 6, 20), 'sigma', bl.oint(0, 2, 20), prior=lambda m, s: 1./s))
T1 = bl.tm.CombinedTransitionModel(bl.tm.GaussianRandomWalk('s1', [0.25, 0.5],
target='mean',
prior=stats.Exponential('e', 0.5)),
bl.tm.GaussianRandomWalk('s2', bl.cint(0, 0.2, 2),
target='sigma',
prior=np.array([0.2, 0.8]))
)
T2 = bl.tm.Independent()
S.addTransitionModel('T1', T1)
S.addTransitionModel('T2', T2)
S.setTransitionModelPrior([0.9, 0.1])
data = np.array([1, 2, 3, 4, 5])
for d in data:
S.step(d)
# test transition model distributions
np.testing.assert_allclose(S.getCurrentTransitionModelDistribution(local=False)[1],
[0.49402616, 0.50597384],
rtol=1e-05, err_msg='Erroneous transition model probabilities.')
np.testing.assert_allclose(S.getCurrentTransitionModelDistribution(local=True)[1],
[0.81739495, 0.18260505],
rtol=1e-05, err_msg='Erroneous local transition model probabilities.')
# test hyper-parameter distributions
np.testing.assert_allclose(S.getCurrentHyperParameterDistribution('s2')[1],
[0.19047162, 0.80952838],
rtol=1e-05, err_msg='Erroneous hyper-parameter distribution.')
# test parameter distributions
np.testing.assert_allclose(S.getParameterDistributions('mean', density=False)[1][:, 5],
[0.05825921, 0.20129444, 0.07273516, 0.02125759, 0.0039255],
rtol=1e-05, err_msg='Erroneous posterior distribution values.')
# test parameter mean values
np.testing.assert_allclose(S.getParameterMeanValues('mean'),
[1.0771838, 1.71494272, 2.45992376, 3.34160617, 4.39337253],
rtol=1e-05, err_msg='Erroneous posterior mean values.')
# test model evidence value
np.testing.assert_almost_equal(S.logEvidence, -9.46900822686, decimal=5,
err_msg='Erroneous log-evidence value.')
def test_fit_hyperprior_sympy(self):
# carry out fit
S = bl.HyperStudy()
S.loadData(np.array([1, 2, 3, 4, 5]))
S.setOM(
bl.om.Gaussian('mean',
bl.cint(0, 6, 20),
'sigma',
bl.oint(0, 2, 20),
prior=lambda m, s: 1 / s**3))
S.setTM(
bl.tm.GaussianRandomWalk('sigma',
bl.cint(0, 0.2, 2),
target='mean',
prior=stats.Exponential('e', 1.)))
S.fit()
# test parameter distributions
np.testing.assert_allclose(
S.getParameterDistributions('mean', density=False)[1][:, 5],
[0.01012545, 0.00761074, 0.00626273, 0.00568207, 0.00562725],
rtol=1e-05,
err_msg='Erroneous posterior distribution values.')
# test parameter mean values
np.testing.assert_allclose(
S.getParameterMeanValues('mean'),
[2.89548676, 2.93742566, 3., 3.06257434, 3.10451324],
rtol=1e-05,
err_msg='Erroneous posterior mean values.')
# test model evidence value
np.testing.assert_almost_equal(S.logEvidence,
-17.0866290887,
decimal=5,
err_msg='Erroneous log-evidence value.')
# test hyper-parameter distribution
x, p = S.getHyperParameterDistribution('sigma')
np.testing.assert_allclose(
np.array([x, p]), [[0., 0.2], [0.487971, 0.512029]],
rtol=1e-05,
err_msg='Erroneous values in hyper-parameter distribution.')
def exponential_to_sympy(node, input_vars=None, log=False):
result = get_density(st.Exponential("Node%s" % node.id, node.l), node,
input_vars)
if log:
result = sp.log(result)
return result
