def _m_step(
self,
x,
quadratic_form,
affiliation,
saliency,
hermitize,
covariance_norm,
eigenvalue_floor,
weight_constant_axis,
):
weight = estimate_mixture_weight(
affiliation=affiliation,
saliency=saliency,
weight_constant_axis=weight_constant_axis,
)
if saliency is None:
masked_affiliation = affiliation
else:
masked_affiliation = affiliation * saliency[..., None, :]
cacg = ComplexAngularCentralGaussianTrainer()._fit(
y=x[..., None, :, :],
saliency=masked_affiliation,
quadratic_form=quadratic_form,
hermitize=hermitize,
covariance_norm=covariance_norm,
eigenvalue_floor=eigenvalue_floor,
)
return CACGMM(weight=weight, cacg=cacg)
def _m_step(
self,
x,
affiliation,
saliency,
weight_constant_axis,
covariance_type,
fixed_covariance,
):
weight = estimate_mixture_weight(
affiliation=affiliation,
saliency=saliency,
weight_constant_axis=weight_constant_axis,
)
gaussian = GaussianTrainer()._fit(y=x[..., None, :, :],
saliency=affiliation *
saliency[..., None, :],
covariance_type=covariance_type)
if fixed_covariance is not None:
assert fixed_covariance.shape == gaussian.covariance.shape, (
f'{fixed_covariance.shape} != {gaussian.covariance.shape}')
gaussian = gaussian.__class__(mean=gaussian.mean,
covariance=fixed_covariance)
return GMM(weight=weight, gaussian=gaussian)
def _m_step(
self,
y,
affiliation,
saliency,
weight_constant_axis,
min_concentration,
max_concentration,
):
weight = estimate_mixture_weight(
affiliation=affiliation,
saliency=saliency,
weight_constant_axis=weight_constant_axis,
)
vmf = VonMisesFisherTrainer()._fit(
y=y[..., None, :, :],
saliency=affiliation * saliency[..., None, :],
min_concentration=min_concentration,
max_concentration=max_concentration,
)
return VMFMM(weight=weight, vmf=vmf)
def _m_step(
self,
y,
affiliation,
saliency,
weight_constant_axis,
):
weight = estimate_mixture_weight(
affiliation=affiliation,
saliency=saliency,
weight_constant_axis=weight_constant_axis,
)
if saliency is None:
masked_affiliation = affiliation
else:
masked_affiliation = affiliation * saliency[..., None, :]
complex_bingham = self.complex_bingham_trainer._fit(
y=y[..., None, :, :],
saliency=masked_affiliation,
)
return CBMM(weight=weight, complex_bingham=complex_bingham)
def _m_step(
self,
y,
affiliation,
saliency,
weight_constant_axis,
):
weight = estimate_mixture_weight(
affiliation=affiliation,
saliency=saliency,
weight_constant_axis=weight_constant_axis,
# dirichlet_prior_concentration=dirichlet_prior_concentration,
)
if saliency is None:
masked_affiliation = affiliation
else:
masked_affiliation = affiliation * saliency[..., None, :]
complex_watson = self.complex_watson_trainer._fit(
y=y[..., None, :, :],
saliency=masked_affiliation,
)
return CWMM(weight=weight, complex_watson=complex_watson)