def __init__(self, dim=3):
self.a = np.random.random((dim, dim))
self.a = np.matmul(self.a, np.transpose(self.a)) + np.eye(dim)
self.par = vb.ModelParamsDict()
self.par.push_param(vb.VectorParam('x', size=dim, lb=0))
self.par.push_param(vb.VectorParam('y', size=dim, lb=1))
def get_group_parameters(K, num_groups=1):
group_par = vb.ModelParamsDict('Single group GLMM parameters')
group_par.push_param(vb.UVNParam('mu'))
group_par.push_param(vb.GammaParam('tau'))
group_par.push_param(vb.UVNParamVector('beta', K))
group_par.push_param(vb.UVNParamVector('u', num_groups))
return group_par
def get_mixture_parameters(dim, num_components):
mix_par = vb.ModelParamsDict()
mix_par.push_param(
vb.ArrayParam('loc', shape=(num_components, dim)))
mix_par.push_param(vb.PosDefMatrixParamArray(
'info', array_shape=(num_components, ), matrix_size=dim))
mix_par.push_param(vb.SimplexParam('w', shape=(1, num_components)))
return mix_par
def get_mle_parameters(K, NG):
mle_par = vb.ModelParamsDict('GLMER Parameters')
mle_par.push_param(vb.VectorParam('mu'))
mle_par.push_param(vb.VectorParam('tau'))
mle_par.push_param(vb.VectorParam('beta', K))
mle_par.push_param(vb.VectorParam('u', NG))
return mle_par
def __init__(self, name='', size=2, diag_lb=0.0, min_df=None):
super().__init__(name=name)
self.__size = int(size)
if not min_df:
min_df = size - 1
assert min_df >= size - 1
self.push_param(vb.ScalarParam('df', lb=min_df))
self.push_param(vb.PosDefMatrixParam('v', diag_lb=diag_lb))
def get_glmm_parameters(
K, NG,
mu_info_min=0.0, tau_alpha_min=0.0, tau_beta_min=0.0,
beta_diag_min=0.0, u_info_min=0.0):
glmm_par = vb.ModelParamsDict('GLMM Parameters')
glmm_par.push_param(vb.UVNParam('mu', min_info=mu_info_min))
glmm_par.push_param(
vb.GammaParam('tau', min_shape=tau_alpha_min, min_rate=tau_beta_min))
#glmm_par.push_param(vb.MVNParam('beta', K, min_info=beta_diag_min))
glmm_par.push_param(vb.UVNParamVector('beta', K, min_info=beta_diag_min))
glmm_par.push_param(vb.UVNParamVector('u', NG, min_info=u_info_min))
return glmm_par
def test_index_params(self):
dim = 3
param = vb.ModelParamsDict('test')
param.push_param(vb.VectorParam('x', size=dim, lb=-2.0, ub=5.0))
param.push_param(vb.VectorParam('y', size=dim, lb=-2.0, ub=5.0))
index_par = obj_lib.make_index_param(param)
param.set_free(np.random.random(param.free_size()))
param_vec = param.get_vector()
for d in range(dim):
for pname in ['x', 'y']:
self.assertAlmostEqual(param[pname].get()[d],
param_vec[index_par[pname].get()[d]])
def __init__(self, dim):
self.dim = dim
self.x = vb.VectorParam('x', size=dim, lb=-2.0, ub=5.0)
self.y = vb.VectorParam('y', size=dim, lb=-2.0, ub=5.0)
self.a_mat = np.full((dim, dim), 0.1) + np.eye(dim)
# For testing the Jacobian
self.b_mat = self.a_mat[0:(dim - 1), 0:dim]
self.set_inits()
self.opt_x = np.linspace(1., 2., self.dim)
self.preconditioner = np.eye(dim)
def get_default_prior_params(K):
prior_par = vb.ModelParamsDict('Prior Parameters')
prior_par.push_param(
vb.VectorParam('beta_prior_mean', K, val=np.zeros(K)))
prior_par.push_param(
vb.PosDefMatrixParam('beta_prior_info', K, val=0.01 * np.eye(K)))
prior_par.push_param(vb.ScalarParam('mu_prior_mean', val=0))
prior_par.push_param(vb.ScalarParam('mu_prior_info', val=0.5))
prior_par.push_param(vb.ScalarParam('tau_prior_alpha', val=3.0))
prior_par.push_param(vb.ScalarParam('tau_prior_beta', val=10.0))
return prior_par
def __init__(self, glmm_par, global_only=False):
self.glmm_par = glmm_par
self.__global_only = global_only
K = glmm_par['beta']['mean'].size()
NG = glmm_par['u']['mean'].size()
self.moment_par = vb.ModelParamsDict('Moment Parameters')
self.moment_par.push_param(vb.VectorParam('e_beta', K))
self.moment_par.push_param(vb.ScalarParam('e_mu'))
self.moment_par.push_param(vb.ScalarParam('e_tau'))
self.moment_par.push_param(vb.ScalarParam('e_log_tau'))
if not self.__global_only:
self.moment_par.push_param(vb.VectorParam('e_u', NG))
self.get_moment_jacobian = \
autograd.jacobian(self.get_moment_vector_from_free)
def get_global_parameters(K):
global_par = vb.ModelParamsDict('Global GLMM parameters')
global_par.push_param(vb.UVNParam('mu'))
global_par.push_param(vb.GammaParam('tau'))
global_par.push_param(vb.UVNParamVector('beta', K))
return global_par
def __init__(self, name='', min_shape=0.0, min_rate=0.0):
super().__init__(name=name)
self.push_param(vb.ScalarParam('shape', lb=min_shape))
self.push_param(vb.ScalarParam('rate', lb=min_rate))
def __init__(self, name='', shape=(1, 2), min_alpha=0.0, val=None):
super().__init__(name=name)
self.__shape = shape
assert min_alpha >= 0, 'alpha parameter must be non-negative'
self.push_param(
vb.ArrayParam('alpha', shape=shape, lb=min_alpha, val=val))