def test_hawkes_sumgaussians_set_data(self):
"""...Test set_data method of Hawkes SumGaussians
"""
events = [[
np.array([1, 1.2, 3.4, 5.8, 10.3, 11, 13.4]),
np.array([2, 5, 8.3, 9.10, 15, 18, 20, 33])
],
[
np.array([2, 3.2, 11.4, 12.8, 45]),
np.array([2, 3, 8.8, 9, 15.3, 19])
]]
learner = HawkesSumGaussians(1)
learner._set_data(events)
self.assertEqual(learner.n_nodes, 2)
events = [
np.array([1, 1.2, 3.4, 5.8, 10.3, 11, 13.4]),
np.array([2, 5, 8.3, 9.10, 15, 18, 20, 33])
]
learner = HawkesSumGaussians(1)
learner._set_data(events)
self.assertEqual(learner.n_nodes, 2)
msg = "All realizations should have 2 nodes, but realization 1 has " \
"1 nodes"
with self.assertRaisesRegex(RuntimeError, msg):
events = [[
np.array([1, 1.2, 3.4, 5.8, 10.3, 11, 13.4]),
np.array([2, 5, 8.3, 9.10, 15, 18, 20, 33])
], [np.array([2, 3.2, 11.4, 12.8, 45])]]
learner._set_data(events)
def test_hawkes_sumgaussians_C_parameter(self):
"""...Test that hawkes sumgaussians C parameter is correctly linked
"""
# First leaner initialization
lasso_grouplasso_ratio = 0.3
learner = HawkesSumGaussians(
1, C=self.float_1, lasso_grouplasso_ratio=lasso_grouplasso_ratio)
strength_lasso = learner.lasso_grouplasso_ratio / self.float_1
strength_grouplasso = (1. - learner.lasso_grouplasso_ratio) / \
self.float_1
self.assertEqual(learner.strength_lasso, strength_lasso)
self.assertEqual(learner.strength_grouplasso, strength_grouplasso)
self.assertEqual(learner._learner.get_strength_lasso(), strength_lasso)
self.assertEqual(learner._learner.get_strength_grouplasso(),
strength_grouplasso)
self.assertEqual(learner.lasso_grouplasso_ratio,
lasso_grouplasso_ratio)
# Change C
learner.C = self.float_2
strength_lasso = learner.lasso_grouplasso_ratio / self.float_2
strength_grouplasso = (1. - learner.lasso_grouplasso_ratio) / \
self.float_2
self.assertEqual(learner.strength_lasso, strength_lasso)
self.assertEqual(learner.strength_grouplasso, strength_grouplasso)
self.assertEqual(learner._learner.get_strength_lasso(), strength_lasso)
self.assertEqual(learner._learner.get_strength_grouplasso(),
strength_grouplasso)
self.assertAlmostEqual(learner.C, self.float_2)
self.assertEqual(learner.lasso_grouplasso_ratio,
lasso_grouplasso_ratio)
def test_hawkes_sumgaussians_lasso_grouplasso_ratio_parameter(self):
"""...Test that hawkes sumgaussians lasso_grouplasso_ratio parameter is
correctly linked
"""
# First learner initialization
C = 5e-3
learner = HawkesSumGaussians(1,
lasso_grouplasso_ratio=self.float_1,
C=C)
strength_lasso = self.float_1 / learner.C
strength_grouplasso = (1. - self.float_1) / learner.C
self.assertEqual(learner.strength_lasso, strength_lasso)
self.assertEqual(learner.strength_grouplasso, strength_grouplasso)
self.assertEqual(learner._learner.get_strength_lasso(), strength_lasso)
self.assertEqual(learner._learner.get_strength_grouplasso(),
strength_grouplasso)
self.assertEqual(learner.C, C)
# change lasso_grouplasso_ratio
learner.lasso_grouplasso_ratio = self.float_2
strength_lasso = self.float_2 / learner.C
strength_grouplasso = (1. - self.float_2) / learner.C
self.assertEqual(learner.strength_lasso, strength_lasso)
self.assertEqual(learner.strength_grouplasso, strength_grouplasso)
self.assertEqual(learner._learner.get_strength_lasso(), strength_lasso)
self.assertEqual(learner._learner.get_strength_grouplasso(),
strength_grouplasso)
self.assertEqual(learner.lasso_grouplasso_ratio, self.float_2)
self.assertEqual(learner.C, C)
def test_hawkes_sumgaussians_parameters(self):
"""...Test that hawkes sumgaussians parameters are correctly linked
"""
learner = HawkesSumGaussians(1, n_gaussians=self.int_1)
self.assertEqual(learner.n_gaussians, self.int_1)
self.assertEqual(learner._learner.get_n_gaussians(), self.int_1)
learner.n_gaussians = self.int_2
self.assertEqual(learner.n_gaussians, self.int_2)
self.assertEqual(learner._learner.get_n_gaussians(), self.int_2)
learner = HawkesSumGaussians(max_mean_gaussian=self.float_1)
self.assertEqual(learner.max_mean_gaussian, self.float_1)
self.assertEqual(learner._learner.get_max_mean_gaussian(),
self.float_1)
learner.max_mean_gaussian = self.float_2
self.assertEqual(learner.max_mean_gaussian, self.float_2)
self.assertEqual(learner._learner.get_max_mean_gaussian(),
self.float_2)
learner = HawkesSumGaussians(1, step_size=self.float_1)
self.assertEqual(learner.step_size, self.float_1)
self.assertEqual(learner._learner.get_step_size(), self.float_1)
learner.step_size = self.float_2
self.assertEqual(learner.step_size, self.float_2)
self.assertEqual(learner._learner.get_step_size(), self.float_2)
def test_hawkes_sumgaussians_solution(self):
"""...Test solution obtained by HawkesSumGaussians on toy timestamps
"""
events = [[
np.array([1, 1.2, 3.4, 5.8, 10.3, 11, 13.4]),
np.array([2, 5, 8.3, 9.10, 15, 18, 20, 33])
],
[
np.array([2, 3.2, 11.4, 12.8, 45]),
np.array([2, 3, 8.8, 9, 15.3, 19])
]]
n_nodes = len(events[0])
n_gaussians = 3
max_mean_gaussian = 5
step_size = 1e-3
C = 10
lasso_grouplasso_ratio = 0.7
baseline_start = np.zeros(n_nodes) + .2
amplitudes_start = np.zeros((n_nodes, n_nodes, n_gaussians)) + .2
learner = HawkesSumGaussians(
n_gaussians=n_gaussians,
max_mean_gaussian=max_mean_gaussian,
step_size=step_size,
C=C,
lasso_grouplasso_ratio=lasso_grouplasso_ratio,
n_threads=3,
max_iter=10,
verbose=False,
em_max_iter=3)
learner.fit(events[0],
baseline_start=baseline_start,
amplitudes_start=amplitudes_start)
baseline = np.array([0.0979586, 0.15552228])
amplitudes = np.array([[[0.20708954, -0.00627318, 0.08388442],
[-0.00341803, 0.34805652, -0.00687372]],
[[-0.00341635, 0.1608013, 0.05531324],
[-0.00342652, -0.00685425, 0.19046195]]])
np.testing.assert_array_almost_equal(learner.baseline,
baseline,
decimal=6)
np.testing.assert_array_almost_equal(learner.amplitudes,
amplitudes,
decimal=6)
kernel_values = np.array([
-0.00068796, 0.01661161, 0.08872543, 0.21473618, 0.25597692,
0.15068586, 0.04194497, 0.00169372, -0.00427233, -0.00233042
])
kernels_norm = np.array([[0.28470077, 0.33776477],
[0.21269818, 0.18018118]])
np.testing.assert_almost_equal(
learner.get_kernel_values(0, 1, np.linspace(0, 4, 10)),
kernel_values)
np.testing.assert_almost_equal(learner.get_kernel_norms(),
kernels_norm)
means_gaussians = np.array([0., 1.66666667, 3.33333333])
std_gaussian = 0.5305164769729844
np.testing.assert_array_almost_equal(learner.means_gaussians,
means_gaussians)
self.assertEqual(learner.std_gaussian, std_gaussian)
learner.n_gaussians = learner.n_gaussians + 1
means_gaussians = np.array([0., 1.25, 2.5, 3.75])
std_gaussian = 0.3978873577297384
np.testing.assert_array_almost_equal(learner.means_gaussians,
means_gaussians)
self.assertEqual(learner.std_gaussian, std_gaussian)
from tick.inference import HawkesSumGaussians
end_time = 1000
n_nodes = 2
n_realizations = 10
n_gaussians = 5
timestamps_list = []
kernel_timefunction = HawkesKernelTimeFunc(
t_values=np.array([0., .7, 2.5, 3., 4.]),
y_values=np.array([.3, .03, .03, .2, 0.]))
kernels = [[HawkesKernelExp(.2, 2.),
HawkesKernelPowerLaw(.2, .5, 1.3)],
[HawkesKernel0(), kernel_timefunction]]
hawkes = SimuHawkes(baseline=[.5, .2],
kernels=kernels,
end_time=end_time,
verbose=False,
seed=1039)
multi = SimuHawkesMulti(hawkes, n_simulations=n_realizations)
multi.simulate()
learner = HawkesSumGaussians(n_gaussians, max_iter=10)
learner.fit(multi.timestamps)
plot_hawkes_kernels(learner, hawkes=hawkes, support=4)