zoj613/occuspytial

A package for fast bayesian analysis of single season spatial occupancy models.

occuspytial's Issues

Make use of numpy's universal functions to simplify codebase

numpy.add.reduceat can be used instead of the cython module with function

OccuSpytial/occuspytial/gibbs/_cumsum.pyx

Lines 8 to 21 in 5df2425

	def split_sum(double[:] arr, long[:] sections, double[:] out):
	cdef:
	double summation
	long previous, current, size = out.shape[0]
	Py_ssize_t i, j
	with nogil:
	for i in range(size):
	previous = sections[i]
	current = sections[i + 1]
	summation = 0
	for j in range(previous, current):
	summation += arr[j]
	out[i] = summation

Numpy universal functions are fast so the speed should roughly be the same.

Make cython building optional by including the .c files.

This avoids the need to have Cython installed before installation.

Caching values returned by __getitem__ in Data class causes a linear growth in memory

This makes sampling any number of samples greater than ~10k nearly impossible and causes the system to freeze once memory has been full. Maybe caching should be disabled.

Gibbs resample method broken

resample method of gibbs samplers is broken since #16 since the PR removes the _excluded attribute in the PosteriorParameter class.