def test_ppf_cdf_round_trip():
mixture = LogisticMixture.from_samples(
np.array([0.5, 0.4, 0.8, 0.8, 0.9, 0.95, 0.15, 0.1]),
{"num_components": 3})
x = 0.65
prob = mixture.cdf(x)
assert mixture.ppf(prob) == pytest.approx(x, rel=1e-3)
def test_fit_mixture_small():
mixture = LogisticMixture.from_samples(np.array([0.1, 0.2, 0.8, 0.9]),
num_components=2)
for prob in mixture.probs:
assert prob == pytest.approx(0.5, 0.1)
locs = sorted([component.loc for component in mixture.components])
assert locs[0] == pytest.approx(0.15, abs=0.1)
assert locs[1] == pytest.approx(0.85, abs=0.1)
def test_fit_samples(logistic_mixture):
data = np.array([logistic_mixture.sample() for _ in range(0, 1000)])
fitted_mixture = LogisticMixture.from_samples(data, {"num_components": 2})
true_locs = sorted([c.loc for c in logistic_mixture.components])
true_scales = sorted([c.s for c in logistic_mixture.components])
fitted_locs = sorted([c.loc for c in fitted_mixture.components])
fitted_scales = sorted([c.s for c in fitted_mixture.components])
for (true_loc, fitted_loc) in zip(true_locs, fitted_locs):
assert fitted_loc == pytest.approx(float(true_loc), rel=0.2)
for (true_scale, fitted_scale) in zip(true_scales, fitted_scales):
assert fitted_scale == pytest.approx(float(true_scale), rel=0.2)
def test_fit_mixture_large():
data1 = onp.random.logistic(loc=0.7, scale=0.1, size=1000)
data2 = onp.random.logistic(loc=0.4, scale=0.2, size=1000)
data = onp.concatenate([data1, data2])
mixture = LogisticMixture.from_samples(data, num_components=2)
locs = sorted([component.loc for component in mixture.components])
scales = sorted([component.scale for component in mixture.components])
assert locs[0] == pytest.approx(0.4, abs=0.2)
assert locs[1] == pytest.approx(0.7, abs=0.2)
assert scales[0] == pytest.approx(0.1, abs=0.2)
assert scales[1] == pytest.approx(0.2, abs=0.2)
def test_fit_mixture_small(fixed_params):
xscale = Scale(0, 1)
mixture = LogisticMixture.from_samples(
data=np.array([0.1, 0.2, 0.8, 0.9]),
fixed_params=fixed_params,
scale=xscale,
)
for prob in mixture.probs:
assert prob == pytest.approx(0.5, 0.1)
locs = sorted(
[component.base_dist.loc for component in mixture.components])
assert locs[0] == pytest.approx(0.15, abs=0.1)
assert locs[1] == pytest.approx(0.85, abs=0.1)
def test_fit_mixture_large(fixed_params):
xscale = Scale(-2, 3)
data1 = onp.random.logistic(loc=0.7, scale=0.1, size=1000)
data2 = onp.random.logistic(loc=0.4, scale=0.2, size=1000)
data = onp.concatenate([data1, data2])
mixture = LogisticMixture.from_samples(
data=data,
fixed_params=fixed_params,
scale=xscale,
)
# FIXME: What's going on below with scales?
components = sorted([(component.base_dist.loc, component.base_dist.s)
for component in mixture.components])
assert components[0][0] == pytest.approx(xscale.normalize_point(0.4),
abs=0.2)
assert components[1][0] == pytest.approx(xscale.normalize_point(0.7),
abs=0.2)
assert components[0][1] == pytest.approx(0.2, abs=0.2)
assert components[1][1] == pytest.approx(0.1, abs=0.2)
