def test_intercept_only_model(crossed_data):
# using fit
model0 = Model(crossed_data)
model0.fit('Y ~ 1', run=False)
model0.build(backend='pymc3')
model0.fit(samples=1)
# using add
model1 = Model(crossed_data)
model1.add('Y ~ 0')
model1.add('1')
model1.build(backend='pymc3')
model1.fit(samples=1)
# check that fit and add models have same priors for fixed
# effects
priors0 = {
x.name: x.prior.args
for x in model0.terms.values() if not x.random
}
priors1 = {
x.name: x.prior.args
for x in model1.terms.values() if not x.random
}
assert set(priors0) == set(priors1)
def test_distribute_group_specific_effect_over(diabetes_data):
# 163 unique levels of BMI in diabetes_data
# With intercept
model = Model("BP ~ (C(age_grp)|BMI)", diabetes_data)
model.build()
# Treatment encoding because of the intercept
lvls = sorted(list(diabetes_data["age_grp"].unique()))[1:]
assert "C(age_grp)|BMI" in model.terms
assert "1|BMI" in model.terms
assert model.terms["C(age_grp)|BMI"].pymc_coords["C(age_grp)_coord_group_expr"] == lvls
# This is equal to the sub-matrix of Z that corresponds to this term.
# 442 is the number of observations. 163 the number of groups.
# 2 is the number of levels of the categorical variable 'C(age_grp)' after removing
# the reference level. Then the number of columns is 326 = 163 * 2.
assert model.terms["C(age_grp)|BMI"].data.shape == (442, 326)
# Without intercept. Reference level is not removed.
model = Model("BP ~ (0 + C(age_grp)|BMI)", diabetes_data)
model.build()
assert "C(age_grp)|BMI" in model.terms
assert not "1|BMI" in model.terms
assert model.terms["C(age_grp)|BMI"].data.shape == (442, 489)
def test_cell_means_with_covariate(crossed_data):
# build model using formula
model0 = Model(crossed_data)
model0.fit('Y ~ 0 + threecats + continuous', run=False)
model0.build()
model0.fit(samples=1)
# build model using add_term
model1 = Model(crossed_data)
model1.add_y('Y')
model1.add_term('threecats', drop_first=False)
model1.add_term('continuous')
model1.build()
model1.fit(samples=1)
# check that design matries are the same,
# even if term names / level names / order of columns is different
X0 = set([tuple(t.data[:,lev]) for t in model0.fixed_terms.values() for lev in range(len(t.levels))])
X1 = set([tuple(t.data[:,lev]) for t in model1.fixed_terms.values() for lev in range(len(t.levels))])
assert X0 == X1
# check that threecats priors have finite variance
assert not any(np.isinf(model0.terms['threecats'].prior.args['sd']))
# check that add_formula and add_term models have same priors for fixed effects
priors0 = {x.name:x.prior.args for x in model0.terms.values() if not x.random}
priors1 = {x.name:x.prior.args for x in model1.terms.values() if not x.random}
assert set(priors0) == set(priors1)
def test_many_fixed_effects(crossed_data):
# build model using formula
model0 = Model(crossed_data)
model0.fit('Y ~ continuous + dummy + threecats', run=False)
model0.build()
model0.fit(samples=1)
# build model using add_term
model1 = Model(crossed_data)
model1.add_y('Y')
model1.add_intercept()
model1.add_term('continuous')
model1.add_term('dummy')
model1.add_term('threecats')
model1.build()
model1.fit(samples=1)
# check that term names agree
assert set(model0.term_names) == set(model1.term_names)
# check that design matries are the same,
# even if term names / level names / order of columns is different
X0 = set([tuple(t.data[:,lev]) for t in model0.fixed_terms.values() for lev in range(len(t.levels))])
X1 = set([tuple(t.data[:,lev]) for t in model1.fixed_terms.values() for lev in range(len(t.levels))])
assert X0 == X1
# check that add_formula and add_term models have same priors for fixed effects
priors0 = {x.name:x.prior.args for x in model0.terms.values() if not x.random}
priors1 = {x.name:x.prior.args for x in model1.terms.values() if not x.random}
assert set(priors0) == set(priors1)
def test_cell_means_with_random_intercepts(crossed_data):
# using formula
model0 = Model(crossed_data)
model0.fit('Y ~ 0 + threecats', random=['subj'], run=False)
model0.build()
model0.fit(samples=1)
# using add_term
model1 = Model(crossed_data, intercept=False)
model1.add_y('Y')
model1.add_term('threecats', categorical=True, drop_first=False)
model1.add_term('subj', categorical=True, random=True, drop_first=False)
model1.build()
model1.fit(samples=1)
# check that they have the same random terms
assert set(model0.random_terms) == set(model1.random_terms)
# check that fixed effect design matries are the same,
# even if term names / level names / order of columns is different
X0 = set([tuple(t.data[:,lev]) for t in model0.fixed_terms.values() for lev in range(len(t.levels))])
X1 = set([tuple(t.data[:,lev]) for t in model1.fixed_terms.values() for lev in range(len(t.levels))])
assert X0 == X1
# check that add_formula and add_term models have same priors for fixed effects
priors0 = {x.name:x.prior.args for x in model0.terms.values() if not x.random}
priors1 = {x.name:x.prior.args for x in model1.terms.values() if not x.random}
assert set(priors0) == set(priors1)
# check that add_formula and add_term models have same priors for random effects
priors0 = {x.name:x.prior.args['sd'].args for x in model0.terms.values() if x.random}
priors1 = {x.name:x.prior.args['sd'].args for x in model1.terms.values() if x.random}
assert set(priors0) == set(priors1)
def test_one_shot_formula_fit(diabetes_data):
model = Model(diabetes_data)
model.fit('S3 ~ S1 + S2', samples=50, run=False)
model.build(backend='pymc3')
nv = model.backend.model.named_vars
targets = ['S3', 'S1', 'Intercept']
assert len(set(nv.keys()) & set(targets)) == 3
def test_cell_means_parameterization(crossed_data):
# build model using fit
model0 = Model(crossed_data)
model0.fit("Y ~ 0 + threecats", run=False)
model0.build(backend="pymc3")
model0.fit(tune=0, samples=1, init=None)
# build model using add
model1 = Model(crossed_data)
model1.add("Y ~ 0")
model1.add("0 + threecats")
model1.build(backend="pymc3")
model1.fit(tune=0, samples=1)
# check that design matrices are the same,
# even if term names / level names / order of columns is different
X0 = set(
[tuple(t.data[:, lev]) for t in model0.fixed_terms.values() for lev in range(len(t.levels))]
)
X1 = set(
[tuple(t.data[:, lev]) for t in model1.fixed_terms.values() for lev in range(len(t.levels))]
)
assert X0 == X1
# check that fit and add models have same priors for fixed
# effects
priors0 = {x.name: x.prior.args for x in model0.terms.values() if not x.random}
priors1 = {x.name: x.prior.args for x in model1.terms.values() if not x.random}
assert set(priors0) == set(priors1)
def test_one_shot_formula_fit(diabetes_data):
model = Model(diabetes_data)
model.fit("S3 ~ S1 + S2", samples=50, run=False)
model.build(backend="pymc3")
nv = model.backend.model.named_vars
targets = ["S3", "S1", "Intercept"]
assert len(set(nv.keys()) & set(targets)) == 3
def test_model_term_names_property(diabetes_data):
model = Model(diabetes_data)
model.add("BMI ~ age_grp")
model.add("BP")
model.add("S1")
model.build(backend="pymc")
assert model.term_names == ["Intercept", "age_grp", "BP", "S1"]
def test_plot_priors(crossed_data):
model = Model("Y ~ 0 + threecats", crossed_data)
# Priors cannot be plotted until model is built.
with pytest.raises(ValueError):
model.plot_priors()
model.build()
model.plot_priors()
def test_model_graph(crossed_data):
model = Model("Y ~ 0 + threecats", crossed_data)
# Graph cannot be plotted until model is built.
with pytest.raises(ValueError):
model.graph()
model.build()
model.graph()
def test_auto_scale(diabetes_data):
# By default, should scale everything except custom Prior() objects
model = Model(diabetes_data)
priors = {"S1": 0.3, "BP": Prior("Cauchy", alpha=1, beta=17.5)}
model.fit("BMI ~ S1 + S2 + BP", run=False, priors=priors)
model.build(backend="pymc3")
p1 = model.terms["S1"].prior
p2 = model.terms["S2"].prior
p3 = model.terms["BP"].prior
assert p1.name == p2.name == "Normal"
assert 0 < p1.args["sd"] < 1
assert p2.args["sd"] > p1.args["sd"]
assert p3.name == "Cauchy"
assert p3.args["beta"] == 17.5
# With auto_scale off, everything should be flat unless explicitly named
# in priors
model = Model(diabetes_data, auto_scale=False)
model.fit("BMI ~ S1 + S2 + BP", run=False, priors=priors)
model.build(backend="pymc3")
p1_off = model.terms["S1"].prior
p2_off = model.terms["S2"].prior
p3_off = model.terms["BP"].prior
assert p1_off.name == "Normal"
assert p2_off.name == "Flat"
assert 0 < p1_off.args["sd"] < 1
assert "sd" not in p2_off.args
assert p3_off.name == "Cauchy"
assert p3_off.args["beta"] == 17.5
def test_slope_only_model(crossed_data):
# using fit
model0 = Model(crossed_data)
model0.fit('Y ~ 0 + continuous', run=False)
model0.build(backend='pymc3')
model0.fit(tune=0, samples=1, init=None)
# using add
model1 = Model(crossed_data)
model1.add('Y ~ 0')
model1.add('0 + continuous')
model1.build(backend='pymc3')
model1.fit(tune=0, samples=1)
# check that term names agree
assert set(model0.term_names) == set(model1.term_names)
# check that fit and add models have same priors for fixed
# effects
priors0 = {
x.name: x.prior.args
for x in model0.terms.values() if not x.random
}
priors1 = {
x.name: x.prior.args
for x in model1.terms.values() if not x.random
}
assert set(priors0) == set(priors1)
def test_model_term_names_property(diabetes_data):
model = Model(diabetes_data)
model.add('BMI ~ age_grp')
model.add('BP')
model.add('S1')
model.build(backend='pymc')
assert model.term_names == ['Intercept', 'age_grp', 'BP', 'S1']
def test_auto_scale(diabetes_data):
# By default, should scale everything except custom Prior() objects
model = Model(diabetes_data)
priors = {'S1': 0.3, 'BP': Prior('Cauchy', alpha=1, beta=17.5)}
model.fit('BMI ~ S1 + S2 + BP', run=False, priors=priors)
model.build(backend='pymc3')
p1 = model.terms['S1'].prior
p2 = model.terms['S2'].prior
p3 = model.terms['BP'].prior
assert p1.name == p2.name == 'Normal'
assert 0 < p1.args['sd'] < 1
assert p2.args['sd'] > p1.args['sd']
assert p3.name == 'Cauchy'
assert p3.args['beta'] == 17.5
# With auto_scale off, everything should be flat unless explicitly named
# in priors
model = Model(diabetes_data, auto_scale=False)
model.fit('BMI ~ S1 + S2 + BP', run=False, priors=priors)
model.build(backend='pymc3')
p1_off = model.terms['S1'].prior
p2_off = model.terms['S2'].prior
p3_off = model.terms['BP'].prior
assert p1_off.name == 'Normal'
assert p2_off.name == 'Flat'
assert 0 < p1_off.args['sd'] < 1
assert 'sd' not in p2_off.args
assert p3_off.name == 'Cauchy'
assert p3_off.args['beta'] == 17.5
def test_cell_means_with_covariate(crossed_data):
# build model using fit
model0 = Model(crossed_data)
model0.fit("Y ~ 0 + threecats + continuous", run=False)
model0.build(backend="pymc3")
# model0.fit(tune=0, samples=1)
# build model using add
model1 = Model(crossed_data)
model1.add("Y ~ 0")
model1.add("0 + threecats")
model1.add("0 + continuous")
model1.build(backend="pymc3")
# model1.fit(tune=0, samples=1)
# check that design matrices are the same,
# even if term names / level names / order of columns is different
X0 = set(
[tuple(t.data[:, lev]) for t in model0.fixed_terms.values() for lev in range(len(t.levels))]
)
X1 = set(
[tuple(t.data[:, lev]) for t in model1.fixed_terms.values() for lev in range(len(t.levels))]
)
assert X0 == X1
# check that threecats priors have finite variance
assert not any(np.isinf(model0.terms["threecats"].prior.args["sd"]))
# check that fit and add models have same priors for fixed
# effects
priors0 = {x.name: x.prior.args for x in model0.terms.values() if not x.random}
priors1 = {x.name: x.prior.args for x in model1.terms.values() if not x.random}
assert set(priors0) == set(priors1)
def test_model_term_names_property_interaction(crossed_data):
crossed_data["fourcats"] = sum([[x] * 10
for x in ["a", "b", "c", "d"]], list()) * 3
model = Model("Y ~ threecats*fourcats", crossed_data)
model.build()
assert model.term_names == [
"Intercept", "threecats", "fourcats", "threecats:fourcats"
]
def test_many_random_effects(crossed_data):
# build model using formula
model0 = Model(crossed_data)
model0.fit('Y ~ continuous',
random=['0+threecats|subj','continuous|item','dummy|item','threecats|site'], run=False)
model0.build()
# model0.fit(samples=1)
# build model using add_term
model1 = Model(crossed_data)
model1.add_y('Y')
# fixed effects
model1.add_intercept()
model1.add_term('continuous')
# random effects
model1.add_term('threecats', over='subj', drop_first=False, random=True,
categorical=True)
model1.add_term('item', random=True, categorical=True)
model1.add_term('continuous', over='item', random=True)
model1.add_term('dummy', over='item', random=True)
model1.add_term('site', random=True, categorical=True)
model1.add_term('threecats', over='site', random=True, categorical=True)
model1.build()
# model1.fit(samples=1)
# check that the random effects design matrices have the same shape
X0 = pd.concat([pd.DataFrame(t.data) if not isinstance(t.data, dict) else
pd.concat([pd.DataFrame(t.data[x]) for x in t.data.keys()], axis=1)
for t in model0.random_terms.values()], axis=1)
X1 = pd.concat([pd.DataFrame(t.data) if not isinstance(t.data, dict) else
pd.concat([pd.DataFrame(t.data[x]) for x in t.data.keys()], axis=1)
for t in model0.random_terms.values()], axis=1)
assert X0.shape == X1.shape
# check that the random effect design matrix contain the same columns,
# even if term names / columns names / order of columns is different
X0_set = set(tuple(X0.iloc[:,i]) for i in range(len(X0.columns)))
X1_set = set(tuple(X1.iloc[:,i]) for i in range(len(X1.columns)))
assert X0_set == X1_set
# check that fixed effect design matries are the same,
# even if term names / level names / order of columns is different
X0 = set([tuple(t.data[:,lev]) for t in model0.fixed_terms.values() for lev in range(len(t.levels))])
X1 = set([tuple(t.data[:,lev]) for t in model1.fixed_terms.values() for lev in range(len(t.levels))])
assert X0 == X1
# check that add_formula and add_term models have same priors for fixed effects
priors0 = {x.name:x.prior.args for x in model0.terms.values() if not x.random}
priors1 = {x.name:x.prior.args for x in model1.terms.values() if not x.random}
assert set(priors0) == set(priors1)
# check that add_formula and add_term models have same priors for random effects
priors0 = {x.name:x.prior.args['sd'].args for x in model0.terms.values() if x.random}
priors1 = {x.name:x.prior.args['sd'].args for x in model1.terms.values() if x.random}
assert set(priors0) == set(priors1)
def test_add_formula_append(diabetes_data):
model = Model(diabetes_data)
model.add('S3 ~ 0')
model.add('S1')
model.build(backend='pymc')
assert hasattr(model, 'y') and model.y is not None and model.y.name == 'S3'
assert 'S1' in model.terms
model.add('S2', append=False)
assert model.y is None
model.add('S3 ~ 0')
model.build(backend='pymc')
assert 'S2' in model.terms
assert 'S1' not in model.terms
def test_add_formula_append(diabetes_data):
model = Model(diabetes_data)
model.add("S3 ~ 0")
model.add("S1")
model.build(backend="pymc")
assert hasattr(model, "y") and model.y is not None and model.y.name == "S3"
assert "S1" in model.terms
model.add("S2", append=False)
assert model.y is None
model.add("S3 ~ 0")
model.build(backend="pymc")
assert "S2" in model.terms
assert "S1" not in model.terms
def test_distribute_random_effect_over(diabetes_data):
# Random slopes
model = Model(diabetes_data)
model.add('BP ~ 1')
model.add(random='C(age_grp)|BMI')
model.build(backend='pymc')
assert model.terms['C(age_grp)[T.1]|BMI'].data.shape == (442, 163)
# Nested or crossed random intercepts
model.reset()
model.add('BP ~ 1')
model.add(random='0+C(age_grp)|BMI')
model.build(backend='pymc')
assert model.terms['C(age_grp)[0]|BMI'].data.shape == (442, 163)
def test_model_terms_cleaned_levels_interaction(crossed_data):
crossed_data["fourcats"] = sum([[x] * 10
for x in ["a", "b", "c", "d"]], list()) * 3
model = Model("Y ~ threecats*fourcats", crossed_data)
model.build()
assert model.terms["threecats:fourcats"].cleaned_levels == [
"threecats[b]:fourcats[b]",
"threecats[b]:fourcats[c]",
"threecats[b]:fourcats[d]",
"threecats[c]:fourcats[b]",
"threecats[c]:fourcats[c]",
"threecats[c]:fourcats[d]",
]
def test_empty_model(crossed_data):
model0 = Model(crossed_data)
model0.add("Y ~ 0")
model0.build(backend="pymc3")
model0.fit(tune=0, samples=1)
model1 = Model(crossed_data)
model1.fit("Y ~ 0", run=False)
model1.build(backend="pymc3")
model1.fit(tune=0, samples=1)
# check that both models have same priors for fixed effects
priors0 = {x.name: x.prior.args for x in model0.terms.values() if not x.random}
priors1 = {x.name: x.prior.args for x in model1.terms.values() if not x.random}
assert set(priors0) == set(priors1)
def test_categorical_term():
data = pd.DataFrame({
"y": np.random.normal(size=6),
"x1": np.random.normal(size=6),
"x2": [1, 1, 0, 0, 1, 1],
"g1": ["a"] * 3 + ["b"] * 3,
"g2": ["x", "x", "z", "z", "y", "y"],
})
model = Model("y ~ x1 + x2 + g1 + (g1|g2) + (x2|g2)", data)
model.build()
terms = ["x1", "x2", "g1", "1|g2", "g1[b]|g2", "x2|g2"]
expecteds = [False, False, True, False, True, False]
for term, expected in zip(terms, expecteds):
assert model.terms[term].categorical is expected
def test_derived_term_search(diabetes_data):
model = Model(diabetes_data)
model.add('BMI ~ 1', random='age_grp|BP', categorical=['age_grp'])
model.build(backend='pymc')
terms = model._match_derived_terms('age_grp|BP')
names = set([t.name for t in terms])
assert names == {'1|BP', 'age_grp[T.1]|BP', 'age_grp[T.2]|BP'}
term = model._match_derived_terms('1|BP')[0]
assert term.name == '1|BP'
# All of these should find nothing
assert model._match_derived_terms('1|ZZZ') is None
assert model._match_derived_terms('ZZZ|BP') is None
assert model._match_derived_terms('BP') is None
assert model._match_derived_terms('BP') is None
def test_derived_term_search(diabetes_data):
model = Model(diabetes_data)
model.add("BMI ~ 1", random="age_grp|BP", categorical=["age_grp"])
model.build(backend="pymc")
terms = model._match_derived_terms("age_grp|BP")
names = set([t.name for t in terms])
assert names == {"1|BP", "age_grp[T.1]|BP", "age_grp[T.2]|BP"}
term = model._match_derived_terms("1|BP")[0]
assert term.name == "1|BP"
# All of these should find nothing
assert model._match_derived_terms("1|ZZZ") is None
assert model._match_derived_terms("ZZZ|BP") is None
assert model._match_derived_terms("BP") is None
assert model._match_derived_terms("BP") is None
def test_3x4_fixed_anova(crossed_data):
# add a four-level category that's perfectly crossed with threecats
crossed_data["fourcats"] = sum([[x] * 10 for x in ["a", "b", "c", "d"]], list()) * 3
# using fit, with intercept
model0 = Model(crossed_data)
model0.fit("Y ~ threecats*fourcats", run=False)
model0.build(backend="pymc3")
fitted0 = model0.fit(tune=0, samples=1, init=None)
assert len(fitted0.posterior.data_vars) == 5
# using fit, without intercept (i.e., 2-factor cell means model)
model1 = Model(crossed_data)
model1.fit("Y ~ 0 + threecats*fourcats", run=False)
model1.build(backend="pymc3")
fitted1 = model1.fit(tune=0, samples=1)
assert len(fitted1.posterior.data_vars) == 4
def test_empty_model(crossed_data):
# using formula
model0 = Model(crossed_data)
model0.add_y('Y')
model0.build()
model0.fit(samples=1)
# using add_term
model1 = Model(crossed_data)
model1.fit('Y ~ 0', run=False)
model1.build()
model1.fit(samples=1)
# check that add_formula and add_term models have same priors for fixed effects
priors0 = {x.name:x.prior.args for x in model0.terms.values() if not x.random}
priors1 = {x.name:x.prior.args for x in model1.terms.values() if not x.random}
assert set(priors0) == set(priors1)
def test_nan_handling(crossed_data):
data = crossed_data.copy()
# Should fail because predictor has NaN
model_fail_na = Model(crossed_data)
model_fail_na.fit('Y ~ continuous', run=False)
model_fail_na.terms['continuous'].data[[4, 6, 8], :] = np.nan
with pytest.raises(ValueError):
model_fail_na.build(backend='pymc3')
# Should drop 3 rows with warning
model_drop_na = Model(crossed_data, dropna=True)
model_drop_na.fit('Y ~ continuous', run=False)
model_drop_na.terms['continuous'].data[[4, 6, 8], :] = np.nan
with pytest.warns(UserWarning) as w:
model_drop_na.build(backend='pymc3')
assert '3 rows' in w[0].message.args[0]
def test_distribute_group_specific_effect_over(diabetes_data):
# 163 unique levels of BMI in diabetes_data
# With intercept
model = Model("BP ~ (C(age_grp)|BMI)", diabetes_data)
model.build()
# Since intercept is present, it uses treatment encoding
lvls = sorted(list(diabetes_data["age_grp"].unique()))[1:]
for lvl in lvls:
assert model.terms[f"C(age_grp)[{lvl}]|BMI"].data.shape == (442, 163)
assert "1|BMI" in model.terms
# Without intercept
model = Model("BP ~ (0 + C(age_grp)|BMI)", diabetes_data)
model.build()
assert model.terms["C(age_grp)[0]|BMI"].data.shape == (442, 163)
assert model.terms["C(age_grp)[1]|BMI"].data.shape == (442, 163)
assert model.terms["C(age_grp)[2]|BMI"].data.shape == (442, 163)
assert not "1|BMI" in model.terms
