def test_particles_no_parameters(transition: Transition):
df = pd.DataFrame(index=[0, 1, 2, 3])
assert len(df) == 4
w = np.array([1, 1, 1, 1]) / 4
transition.fit(df, w)
with pytest.raises(NotEnoughParticles):
transition.required_nr_samples(.1)
def test_variance_no_side_effect(transition: Transition):
df, w = data(60)
transition.fit(df, w)
# very intrusive test. touches internals of m. not very nice.
X_orig_id = id(transition.X)
transition.mean_cv() # this has to be called here
assert id(transition.X) == X_orig_id
def test_pdf_return_types(transition: Transition):
df, w = data(20)
transition.fit(df, w)
single = transition.pdf(df.iloc[0])
multiple = transition.pdf(df)
assert isinstance(single, float)
assert multiple.shape == (20,)
def test_variance_estimate(transition: Transition):
var_list = []
for n in [20, 250]:
df, w = data(n)
transition.fit(df, w)
var = transition.mean_cv()
var_list.append(var)
assert var_list[0] >= var_list[1]
def test_mean_coefficient_of_variation_sample_not_full_rank(
transition: Transition):
"""
This is a test created after I encountered this kind of bug
"""
n = 13
df = pd.DataFrame({"a": np.ones(n) * 2, "b": np.ones(n)})
w = np.ones(len(df)) / len(df)
transition.fit(df, w)
transition.mean_cv()
def test_variance_estimate_higher_n_than_sample(transition: Transition):
n = 100
df, w = data(n)
transition.fit(df, w)
var_list = []
for n_test in [n, n*4, n*10]:
var = transition.mean_cv(n_test)
var_list.append(var)
for lower, upper in zip(var_list[:-1], var_list[1:]):
assert lower >= upper
def test_argument_order(transition: Transition):
"""
Dataframes passed to the transition kernels are generated from dicts.
Order of parameter names is no guaranteed.
The Transition kernel has to take care of the correct sorting.
"""
df, w = data(20)
transition.fit(df, w)
test = df.iloc[0]
reversed = test[::-1]
# works b/c of even nr of parameters
assert (np.array(test) != np.array(reversed)).all()
assert transition.pdf(test) == transition.pdf(reversed)
def test_variance_estimate_higher_n_than_sample(transition: Transition):
n = 100
df, w = data(n)
transition.fit(df, w)
var_list = []
for n_test in [n, n * 4, n * 10]:
var = transition.mean_cv(n_test)
var_list.append(var)
for lower, upper in zip(var_list[:-1], var_list[1:]):
# add a little buffer to overcome slight random fluctuations
assert lower + 1e-2 >= upper
def test_many_particles_single_par(transition: Transition):
df, w = data_single(20)
transition.fit(df, w)
transition.required_nr_samples(.1)
def test_rvs_return_type(transition: Transition):
df, w = data(20)
transition.fit(df, w)
sample = transition.rvs()
assert (sample.index == pd.Index(["a", "b"])).all()
def test_many_particles(transition: Transition):
df, w = data(20)
transition.fit(df, w)
transition.required_nr_samples(0.1)
def test_two_particles_required_nr_samples(transition: Transition):
# TODO define proper behavior
df, w = data(2)
transition.fit(df, w)
transition.required_nr_samples(.1)
def test_two_particles_fit(transition: Transition):
# TODO define proper behavior
df, w = data(2)
transition.fit(df, w)
def test_single_particle_fit(transition: Transition):
# TODO define proper behavior
df, w = data(1)
transition.fit(df, w)
def test_score(transition: Transition):
df, w = data(20)
transition.fit(df, w)
transition.score(df, w) # just call it
def test_0_particles_fit(transition: Transition):
# TODO define proper behavior
df, w = data(0)
with pytest.raises(NotEnoughParticles):
transition.fit(df, w)
def test_single_particle_required_nr_samples(transition: Transition):
# TODO define proper behavior
df, w = data(1)
transition.fit(df, w)
transition.required_nr_samples(0.1)
