def __init__(self, Y, link, trLen, nLeg=50, nHerm=20, dtstep=0.001):
super(PointProcess, self).__init__()
"""
Class for PointProcess expected log likelihood
Y -- data list with [R][D][Nspikes] trials, neurons, number of spikes
link -- non-linearity module
trLen -- list with length of each trial
nLeg -- number of quadrature nodes to use for Gauss Legendre (to evaluate normalizer)
nHerm -- number of Gauss Hermite quadrature nodes to use for evaluating Gaussian expectations of link function (if not analytic)
"""
self.nTrials = len(trLen) # number of trials
self.nOut = len(Y[0]) # number of output dimensions
self.link = link # link function object
self.trLen = trLen # length of each trial in units of time
self.dtstep = dtstep # discretisation for inference
# Gauss Hermite quadrature nodes and weights
xxH, wwH = gauss_hermite(nHerm)
self.xxHerm = xxH
self.wwHerm = wwH
# Gauss Legendre quadrature nodes and weights
xxL = torch.zeros(self.nTrials, nLeg).type(float_type)
wwL = torch.zeros(self.nTrials, nLeg).type(float_type)
for rr in range(self.nTrials):
xxL[rr, :], wwL[rr, :] = gauss_legendre(nLeg, a=0, b=trLen[rr])
self.xxLeg = xxL # R x n
self.wwLeg = wwL
# arrange data and ensure everything is a pytorch variable
self.arrange_data(Y)
def gauss_hermite_design(N):
"""
Tensor product Gauss-Hermite quadrature points.
:param int[] N: Contains the number of points per dimension in the tensor
product design.
:return: design, N-by-m matrix that contains the design points.
:rtyp: ndarray
"""
Npts = int(np.prod(np.array(N)))
logging.getLogger(__name__).debug('Gauss-Hermite design with {:d} points in {:d} dimensions.'.format(Npts, len(N)))
design = gauss_hermite(N)[0]
return design
def gauss_hermite_design(N):
"""Tensor product Gauss-Hermite quadrature points.
Parameters
----------
N : int[]
contains the number of points per dimension in the tensor product design
Returns
-------
design : ndarray
N-by-m matrix that contains the design points
"""
design = gauss_hermite(N)[0]
return design
def unbounded_design(N):
return gauss_hermite(N)[0]
def gauss_hermite_design(N):
return gauss_hermite(N)[0]
