def time_opt(self):
p, x = [], []
for col in range(1, self.num_points):
p.extend(self.upper_tri_col(col))
x.extend(self.diff_col_times(col))
ker = self.time_kernel_provider(x, p)
return kernels.Reflect1D(ker)
def func_opt(self):
times, probs = self._times_probs_for_func_opt()
ker = kernels.GaussianBase(times)
ker.covariance_matrix = 1
ker.bandwidth = self._h
ker.weights = probs
return kernels.Reflect1D(ker)
def test_Reflect1D():
def kernel(pts):
return np.exp(-pts*pts)
k = testmod.Reflect1D(kernel)
assert k.kernel is kernel
assert k(5) == pytest.approx(2*np.exp(-25))
np.testing.assert_allclose(k([5,7]), 2*np.exp([-25, -49]))
def compute_t_marginal(ker):
"""Return the kernel resulting from integrating out all but the first
coordinate. Handles the case when a cutoff is applied, but does not apply
the cutoff in the calculation!
:param ker: Any of the kernels returned by the providers in this module.
"""
if isinstance(ker, kernels.Reflect1D):
return ValueError("Already a one dimensional kernel!")
already_reflected_axis = None
if isinstance(ker, kernels.ReflectedKernel):
already_reflected_axis = ker.reflected_axis
ker = ker.delegate
if isinstance(ker, _WrapWithCutoff):
ker = ker._ker
k = kernels.marginalise_gaussian_kernel(ker, 2)
k = kernels.marginalise_gaussian_kernel(k, 1)
if already_reflected_axis == 0:
return kernels.Reflect1D(k)
return k
def trigger_time_opt(self):
_, trigs = self._sample_points()
ker = self.trigger_time_provider(trigs[0], None)
return kernels.Reflect1D(ker)
def trigger_time_opt(self):
pts, w = self._cached_trigger_data()
kt = self.trigger_time_provider(pts[0], w)
return kernels.Reflect1D(kt)
def func_opt(self):
times, probs = self._times_probs_for_func_opt()
ker = kernels.GaussianNearestNeighbour(times, self._knn)
ker.weights = probs
return kernels.Reflect1D(ker)
def time_opt(self):
times = self._sample_triggers()[0]
ker = self.time_kernel_provider(times, None)
return kernels.Reflect1D(ker)