def test_dichotomous_bmrs():
# not checking if numbers are correct, but that functionality works
session = Session(
mcmc_iterations=2000,
mcmc_num_chains=4,
mcmc_warmup_fraction=0.25,
)
session.add_dichotomous_data(
dose=[0, 1.96, 5.69, 29.75],
n=[75, 49, 50, 49],
incidence=[5, 1, 3, 14]
)
session.add_models(
dmodels.Logistic(),
)
session.add_bmrs(
dbmr.Extra(bmr=0.1, priors=[1, ]),
dbmr.Added(bmr=0.1, priors=[1, ])
)
session.execute()
session.calculate_bmrs()
# check scaler weight
assert session.models[0].model_weight_scaler == 1.
# check bmd medians are calculated
medians1 = [b.model_average['stats']['p50'] for b in session.bmrs]
medians2 = [17.887098904857172, 18.403145098571546]
assert np.all(np.isclose(medians1, medians2))
def test_threaded():
session = Session(
mcmc_iterations=2000,
mcmc_num_chains=2,
mcmc_warmup_fraction=0.25,
)
session.add_dichotomous_data(dose=[0, 1.96, 5.69, 29.75],
n=[75, 49, 50, 49],
incidence=[5, 1, 3, 14])
session.add_models(
dmodels.Logistic(),
dmodels.LogLogistic(),
dmodels.Logistic(),
dmodels.LogLogistic(),
)
# check that both run
session.execute(pythreads=True)
session.execute(pythreads=False)
def test_bmr_priors():
# Check that modifying BMR priors give expected results.
session = Session(
mcmc_iterations=2000,
mcmc_num_chains=2,
mcmc_warmup_fraction=0.25,
)
session.add_dichotomous_data(
dose=[0, 1.96, 5.69, 29.75],
n=[75, 49, 50, 49],
incidence=[5, 1, 3, 14]
)
session.add_models(
dmodels.Logistic(),
dmodels.Probit(),
)
session.add_bmrs(
dbmr.Extra(bmr=0.1, priors=[0, 1]),
dbmr.Extra(bmr=0.1, priors=[0.5, 0.5]),
dbmr.Extra(bmr=0.1, priors=[3, 1]),
)
session.execute()
session.calculate_bmrs()
# get model weights using a non-informative prior
expected_weights = [0.51190026721838455, 0.48809973278161539]
model_weights = [m.model_weight_scaler for m in session.models]
assert np.isclose(model_weights, expected_weights).all()
# when using a zero weight, posterior should also be zero
bmr = session.bmrs[0]
assert np.isclose(bmr.priors, [0., 1.]).all()
assert np.isclose(bmr.get_model_posterior_weights(session.models),
[0., 1.]).all()
# when using equal weights, posterior should be same as model weights
bmr = session.bmrs[1]
assert np.isclose(bmr.get_model_posterior_weights(session.models),
model_weights).all()
# when using non-equal weights, ensure posterior are different and move
# in the expected direction using the weights. Also check to make sure the
# priors are normalized correctly.
bmr = session.bmrs[2]
assert np.isclose(bmr.priors, [0.75, 0.25]).all()
assert np.isclose(bmr.get_model_posterior_weights(session.models),
[0.70457014, 0.29542986]).all()
def test_basic_dichotomous():
session = Session(
mcmc_iterations=2000,
mcmc_num_chains=4,
mcmc_warmup_fraction=0.25,
)
session.add_dichotomous_data(dose=[0, 1.96, 5.69, 29.75],
n=[75, 49, 50, 49],
incidence=[5, 1, 3, 14])
session.add_models(
dmodels.Logistic(),
dmodels.LogLogistic(),
)
session.execute()
weights = [m.model_weight_scaler for m in session.models]
expected = [0.62553842799869863, 0.37446157200130137]
assert np.all(np.isclose(weights, expected))
assert np.isclose(sum(weights), 1.)
def test_extract_summary_value():
model = dmodels.Logistic()
summary_text = 'Inference for Stan model: anon_model_6ec9f268cf3fd9261d734b7a1232b68a.\n2 chains, each with iter=20000; warmup=10000; thin=1; \npost-warmup draws per chain=10000, total post-warmup draws=20000.\n\n mean se_mean sd 2.5% 25% 50% 75% 97.5% n_eff Rhat\na -1.72 2.7e-3 0.17 -2.06 -1.83 -1.71 -1.61 -1.41 3818.0 1.0\nb 1.13 4.2e-3 0.26 0.63 0.96 1.13 1.3 1.65 3856.0 1.0\nlp__ -65.81 0.02 1.01 -68.5 -66.2 -65.5 -65.08 -64.84 4253.0 1.0\n\nSamples were drawn using NUTS at Sun Aug 7 19:14:37 2016.\nFor each parameter, n_eff is a crude measure of effective sample size,\nand Rhat is the potential scale reduction factor on split chains (at \nconvergence, Rhat=1).' # noqa
result = model.extract_summary_values(summary_text)
expected = {
'a': {
'25%': -1.83,
'97.5%': -1.41,
'Rhat': 1.0,
'50%': -1.71,
'75%': -1.61,
'2.5%': -2.06,
'sd': 0.17,
'n_eff': 3818.0,
'se_mean': 0.0027,
'mean': -1.72
}, # noqa
'b': {
'25%': 0.96,
'97.5%': 1.65,
'Rhat': 1.0,
'50%': 1.13,
'75%': 1.3,
'2.5%': 0.63,
'sd': 0.26,
'n_eff': 3856.0,
'se_mean': 0.0042,
'mean': 1.13
}, # noqa
'lp__': {
'25%': -66.2,
'97.5%': -64.84,
'Rhat': 1.0,
'50%': -65.5,
'75%': -65.08,
'2.5%': -68.5,
'sd': 1.01,
'n_eff': 4253.0,
'se_mean': 0.02,
'mean': -65.81
}, # noqa
}
assert expected == result