'''This file includes multiple ways to reorder the graph nodes, including reverse cholesky, min-fill, and min-degree.'''
import numpy as np
import copy
from scipy.sparse import csc_matrix
import itertools
from ...External import SUITESPARSE_SUPPORT
if SUITESPARSE_SUPPORT:
from sksparse.cholmod import cholesky
__all__ = ['ReverseCholesky', 'MinFill', 'MinDegree']
nax = np.newaxis
[docs]class ReverseCholesky():
def __init__(self):
self.id = 0
[docs] def perm(self, gp):
dim = gp.shape[0]
chol_factor = cholesky(csc_matrix(gp))
perm = chol_factor.P()
# reverse ordering
perm = np.flipud(perm)
# # inverse perm
# inv_perm = [0] * dim
# for i, p in enumerate(perm):
# ip[p] = i
return perm
[docs] def invPerm(self, gp):
dim = gp.shape[0]
# inverse perm
perm = self.perm(gp)
invPerm = [0] * dim
for i, p in enumerate(perm):
invPerm[p] = i
return np.array(invPerm)
[docs]class MinFill():
def __init__(self):
self.id = 1
[docs] def perm(self, gp):
dim = gp.shape[0]
gp_copy = copy.copy(gp)
perm = np.zeros(dim,dtype='int')
nodes_avail = [i for i in range(dim)]
for k in range(dim):
# non_zero_sum = np.zeros(dim);
added_edges = np.zeros(dim);
for i in range(dim):
if i in nodes_avail:
# non_zero_sum[i] = np.sum(gp_copy[i,:] != 0)
gpLower = np.tril(gp_copy)
non_zero_ind = np.where(gpLower[k,:k] != 0)[0]
co_parents = list(itertools.combinations(non_zero_ind,2))
for j in range(len(co_parents)):
row_index = max(co_parents[j])
col_index = min(co_parents[j])
if gp_copy[row_index, col_index] == 0:
added_edges[i] += 1
else:
added_edges[i] = dim + 1 # make sure not the minimum
m, ind = min((s,ind) for ind,s in enumerate(added_edges))
# set value of permutation vector
perm[k] = ind
nodes_avail.remove(ind)
# if no fill, just set value
if m == 0:
pass#print("ok!")
# else, must fill
else:
#marry parents of removed edge
gpLower = np.tril(gp_copy)
non_zero_ind = np.where(gpLower[ind,:ind] != 0)[0]
co_parents = list(itertools.combinations(non_zero_ind,2))
for j in range(len(co_parents)):
row_index = max(co_parents[j])
col_index = min(co_parents[j])
gp_copy[row_index, col_index] = 1.0
gp_copy[col_index, row_index] = 1.0
perm = np.flipud(perm)
perm = np.ndarray.tolist(perm)
return np.array(perm)
[docs] def invPerm(self, gp):
dim = gp.shape[0]
# inverse perm
perm = self.perm(gp)
invPerm = [0] * dim
for i, p in enumerate(perm):
invPerm[p] = i
return np.array(invPerm)
[docs]class MinDegree():
def __init__(self):
self.id = 2
[docs] def perm(self, gp):
dim = gp.shape[0]
gp_copy = copy.copy(gp)
perm = np.zeros(dim,dtype='int')
nodes_avail = [i for i in range(dim)]
for k in range(dim):
non_zero_sum = np.zeros(dim);
for i in range(dim):
if i in nodes_avail:
non_zero_sum[i] = np.sum(gp_copy[i,:] != 0)
else:
non_zero_sum[i] = dim + 1 # make sure not the minimum
m, ind = min((s,ind) for ind,s in enumerate(non_zero_sum))
#set value of permutation vector
perm[k] = ind
#marry parents of removed edge
gpLower = np.tril(gp_copy)
non_zero_ind = np.where(gpLower[k,:k] != 0)[0]
if len(non_zero_ind) > 1:
co_parents = list(itertools.combinations(non_zero_ind,2))
for j in range(len(co_parents)):
row_index = max(co_parents[j])
col_index = min(co_parents[j])
gp_copy[row_index, col_index] = 1.0
gp_copy[col_index, row_index] = 1.0
nodes_avail.remove(ind)
perm = np.flipud(perm)
perm = np.ndarray.tolist(perm)
return np.array(perm)
[docs] def invPerm(self, gp):
dim = gp.shape[0]
# inverse perm
perm = self.perm(gp)
invPerm = [0] * dim
for i, p in enumerate(perm):
invPerm[p] = i
return np.array(invPerm)