def test_cached_support(self, rng):
x = rng.normal(0, 3, 100)
kde = KDE()
kde.define_support(x)
_, support = kde(x[(x > -1) & (x < 1)])
assert_array_equal(support, kde.support)
def test_bw_adjust(self, rng):
x = rng.normal(0, 3, 100)
kde1 = KDE(bw_adjust=.2)
kde2 = KDE(bw_adjust=2)
d1, _ = kde1(x)
d2, _ = kde2(x)
assert np.abs(np.diff(d1)).mean() > np.abs(np.diff(d2)).mean()
def test_cumulative(self, rng):
x = rng.normal(0, 3, 1000)
kde = KDE(cumulative=True)
density, _ = kde(x)
assert density[0] == pytest.approx(0, abs=1e-5)
assert density[-1] == pytest.approx(1, abs=1e-5)
def test_cut(self, rng):
x = rng.normal(0, 3, 1000)
kde = KDE(cut=0)
_, support = kde(x)
assert support.min() == x.min()
assert support.max() == x.max()
cut = 2
bw_scale = .5
bw = x.std() * bw_scale
kde = KDE(cut=cut, bw_method=bw_scale, gridsize=1000)
_, support = kde(x)
assert support.min() == pytest.approx(x.min() - bw * cut, abs=1e-2)
assert support.max() == pytest.approx(x.max() + bw * cut, abs=1e-2)
def test_bivariate_cumulative(self, rng):
x, y = rng.normal(0, 3, (2, 50))
kde = KDE(gridsize=100, cumulative=True)
density, _ = kde(x, y)
assert density[0, 0] == pytest.approx(0, abs=1e-2)
assert density[-1, -1] == pytest.approx(1, abs=1e-2)
def test_clip(self, rng):
x = rng.normal(0, 3, 100)
clip = -1, 1
kde = KDE(clip=clip)
_, support = kde(x)
assert support.min() >= clip[0]
assert support.max() <= clip[1]
def test_gridsize(self, rng):
x = rng.normal(0, 3, 1000)
n = 200
kde = KDE(gridsize=n)
density, support = kde(x)
assert density.size == n
assert support.size == n
def test_bivariate_grid(self, rng):
n = 100
x, y = rng.normal(0, 3, (2, 50))
kde = KDE(gridsize=n)
density, (xx, yy) = kde(x, y)
assert density.shape == (n, n)
assert xx.size == n
assert yy.size == n
def test_bivariate_normalization(self, rng):
x, y = rng.normal(0, 3, (2, 50))
kde = KDE(gridsize=100)
density, (xx, yy) = kde(x, y)
dx = xx[1] - xx[0]
dy = yy[1] - yy[0]
total = density.sum() * (dx * dy)
assert total == pytest.approx(1, abs=1e-2)
def get_proba_density_estimation(scores, classes, n_partitions=1000):
# from sklearn.neighbors import KernelDensity
probability_density = {}
from seaborn._statistics import KDE
for i, class_ in enumerate(classes):
selected_proba = np.array(scores[:, i])
selected_proba_series = pd.Series(selected_proba).dropna(
) # todo use numpy instead to remove pandas
# selected_proba = selected_proba.reshape((selected_proba.shape[0], 1))
estimator = KDE(bw_method='scott',
bw_adjust=0.01,
gridsize=200,
cut=3,
clip=None,
cumulative=False)
density, support = estimator(selected_proba_series, weights=None)
probability_density[class_] = {
'gaussian': {
"X": support.tolist(),
"probaDensity": density.tolist()
}
}
return probability_density
def test_density_normalization(self, rng):
x = rng.normal(0, 3, 1000)
kde = KDE()
density, support = kde(x)
assert self.integrate(density, support) == pytest.approx(1, abs=1e-5)