def __init__(self, shape, probc, meanc, varc,name=None,\
var_axes=(0,),mean_fix=None,var_fix=None,\
is_complex=False,map_est=False,zvarmin=1e-3, tune_gmm=False):
# Save properties
self.map_est = map_est
self.is_complex = is_complex
self.zvarmin = zvarmin
self.tune_gmm = tune_gmm
self.mean_fix = mean_fix
self.var_fix = var_fix
dtype = meanc.dtype
BaseEst.__init__(self,shape=shape,var_axes=var_axes,dtype=dtype,\
name=name, type_name='GMM', nvars=1, cost_avail=True)
# If the GMM parameters are given, set them now
self.mix = None
self.set_gmm_param(probc, meanc, varc)
self.it = 0
# Set the indicators on which the variance and means are fixed
if self.mean_fix is None:
self.mean_fix = np.zeros(len(probc))
if self.var_fix is None:
self.var_fix = np.zeros(len(probc))
def __init__(self,shape,var_axes=(0,),max_it=10, step_init=1,\
dtype=np.float64,name=None,\
step_max=1,step_min=1e-8,zinit=None,is_complex=False,gtol=1e-3):
BaseEst.__init__(self,shape=shape,var_axes=var_axes,dtype=dtype,\
name=name, type_name='ScalarNL', nvars=1, cost_avail=True)
# Save parameters
self.max_it = max_it
self.cost_avail = True
self.step_last = step_init
self.step_max = step_max
self.step_min = step_min
self.is_complex = is_complex
self.gtol = gtol
self.min_it = 1
# Set default value that the Hessian is available
self.hess_avail = True
# Set initial point
if (zinit is None):
self.zlast = self.proj(np.zeros(self.shape))
else:
self.zlast = zinit
def __init__(self, zmean, zvar, shape,name=None,\
var_axes = (0,), zmean_axes='all',\
is_complex=False, map_est=False, tune_zvar=False,\
tune_rvar=False):
if np.isscalar(shape):
shape = (shape,)
if is_complex:
dtype = np.double
else:
dtype = np.complex
BaseEst.__init__(self,shape=shape,dtype=dtype,name=name,
var_axes=var_axes,type_name='GaussEst', cost_avail=True)
self.zmean = zmean
self.zvar = zvar
self.cost_avail = True
self.is_complex = is_complex
self.map_est = map_est
self.zmean_axes = zmean_axes
self.tune_zvar = tune_zvar
self.tune_rvar = tune_rvar
ndim = len(self.shape)
if self.zmean_axes == 'all':
self.zmean_axes = tuple(range(ndim))
# If zvar is a scalar, then repeat it to the required shape,
# which are all the dimensions not being averaged over
if np.isscalar(self.zvar):
var_shape = get_var_shape(self.shape, self.var_axes)
self.zvar = np.tile(self.zvar, var_shape)
def __init__(self, zval, pz, shape, var_axes=(0,),\
is_complex=False,name=None):
if np.isscalar(shape):
shape = (shape,)
# Convert scalars to arrays
if np.isscalar(zval):
zval = np.array([zval])
if not is_complex and np.any(zval.imag != 0):
import warnings
warnings.warn('zval is complex, but is_complex is False (forcing True)')
is_complex = True
if np.isscalar(pz):
pz = np.array([pz])
# Set parameters of base estimator
dtype = zval.dtype
BaseEst.__init__(self,shape=shape, var_axes=var_axes, dtype=dtype, name=name,\
type_name='DiscreteEst', nvars=1, cost_avail=True)
# Set parameters
self.zval = zval
self.pz = pz
self.shape = shape
self.is_complex = is_complex
self.fz = -np.log(pz)
def __init__(self,y,shape,var_axes=(0,),thresh=0,perr=1e-6,\
name=None,var_init=np.Inf,dtype=np.float64):
BaseEst.__init__(self, shape=shape, var_axes=var_axes, dtype=dtype,\
name=name,type_name='BinaryQuantEst', nvars=1, cost_avail=True)
self.y = y
self.shape = shape
self.thresh = thresh
self.perr = perr
self.cost_avail = True
self.var_init = var_init
def __init__(self,A,y,wvar=0,\
wrep_axes='all', var_axes=(0,),name=None,map_est=False,\
is_complex=False,rvar_init=1e5,tune_wvar=False):
BaseEst.__init__(self, shape=A.shape0, var_axes=var_axes,\
dtype=A.dtype0, name=name,\
type_name='LinEstim', nvars=1, cost_avail=True)
self.A = A
self.y = y
self.wvar = wvar
self.map_est = map_est
self.is_complex = is_complex
self.cost_avail = True
self.rvar_init = rvar_init
self.tune_wvar = tune_wvar
# Get the input and output shape
self.zshape = A.shape0
self.yshape = A.shape1
# Set the repetition axes
ndim = len(self.yshape)
if wrep_axes == 'all':
wrep_axes = tuple(range(ndim))
self.wrep_axes = wrep_axes
# Compute the SVD terms
# Take an SVD A=USV'. Then write p = SV'z + w,
if not A.svd_avail:
raise common.VpException("Transform must support an SVD")
self.p = A.UsvdH(y)
srep_axes = A.get_svd_diag()[2]
# Compute the norm of ||y-UU*(y)||^2/wvar
if np.all(self.wvar > 0):
yp = A.Usvd(self.p)
wvar1 = common.repeat_axes(wvar,
self.yshape,
self.wrep_axes,
rep=False)
err = np.abs(y - yp)**2
self.ypnorm = np.sum(err / wvar1)
else:
self.ypnorm = 0
# Check that all axes on which A operates are repeated
for i in range(ndim):
if not (i in self.wrep_axes) and not (i in srep_axes):
raise common.VpException(
"Variance must be constant over output axis")
if not (i in self.var_axes) and not (i in srep_axes):
raise common.VpException(
"Variance must be constant over input axis")
def __init__(self,
shape,
var_axes=[(0, ), (0, )],
name=None,
map_est=False):
self.map_est = map_est
# Initial variances
self.zvar0_init = np.Inf
self.zvar1_init = np.Inf
nvars = 2
dtype = np.float64
BaseEst.__init__(self,shape=[shape,shape], var_axes=var_axes, dtype=dtype, name=name,\
type_name='ReLUEst', nvars=nvars, cost_avail=True)
def __init__(self,A,b,wvar=0,var_axes=[(0,), (0,)], wrep_axes='all',\
name=None,map_est=False,is_complex=False,est_meth='svd',\
nit_cg=100, atol_cg=1e-3, save_stats=False):
self.A = A
self.b = b
self.wvar = wvar
self.map_est = map_est
self.is_complex = is_complex
self.cost_avail = True
# Initial variance. This is large value since the quantities
# are underdetermined
self.zvar0_init = np.Inf
self.zvar1_init = np.Inf
# Set parameters of the base estimator
shape = [A.shape0, A.shape1]
dtype = [A.dtype0, A.dtype1]
nvars = 2
for i in range(nvars):
if var_axes[i] == 'all':
ndim = len(shape[i])
var_axes[i] = tuple(range(ndim))
if wrep_axes == 'all':
ndim = len(shape[1])
wrep_axes = tuple(range(ndim))
self.wrep_axes = wrep_axes
BaseEst.__init__(self,shape=shape, var_axes=var_axes, dtype=dtype, name=name,\
type_name='LinEstTwo', nvars=nvars, cost_avail=True)
# Initialization depending on the estimation method
self.est_meth = est_meth
if self.est_meth == 'svd':
self.init_svd()
elif self.est_meth == 'cg':
self.init_cg()
else:
raise common.VpException(
"Unknown estimation method {0:s}".format(est_meth))
# CG parameters
self.nit_cg = nit_cg
self.atol_cg = atol_cg
self.save_stats = save_stats
self.init_hist_dict()
def __init__(self, est_list, name=None):
# Save estimator list
self.est_list = est_list
# Get parameters from the estimators
shape = []
var_axes = []
dtype = []
cost_avail = True
nvars = len(est_list)
for est in est_list:
shape.append(est.shape)
var_axes.append(est.var_axes)
dtype.append(est.dtype)
cost_avail = cost_avail and est.cost_avail
self.cost_avail = cost_avail
BaseEst.__init__(self,shape=shape,var_axes=shape,dtype=dtype,\
name=name, type_name='StackEst', nvars=nvars, cost_avail=cost_avail)
def __init__(self, zval, pz, shape, var_axes=(0,),\
is_complex=False,name=None):
# Convert scalars to arrays
if np.isscalar(zval):
zval = np.array([zval])
if np.isscalar(pz):
pz = np.array([pz])
# Set parameters of base estimator
dtype = zval.dtype
BaseEst.__init__(self,shape=shape, var_axes=var_axes, dtype=dtype, name=name,\
type_name='DiscreteEst', nvars=1, cost_avail=True)
# Set parameters
self.zval = zval
self.pz = pz
self.shape = shape
self.is_complex = is_complex
self.fz = -np.log(pz)
def __init__(self, est_list, w, name=None):
self.est_list = est_list
self.w = w
shape = est_list[0].shape
var_axes = est_list[0].var_axes
dtype = est_list[0].dtype
# Check that all estimators have cost available
for est in est_list:
if est.shape != shape:
raise common.VpException('Estimators must have the same shape')
if est.var_axes != var_axes:
raise common.VpException(
'Estimators must have the same var_axes')
if not est.cost_avail:
raise common.VpException(\
"Estimators in a mixture distribution"+\
"must have cost_avail==True")
BaseEst.__init__(self,shape=shape,var_axes=var_axes,dtype=dtype,\
name=name, type_name='Mixture', nvars=1, cost_avail=True)