def algo_head_tail_bisearch(edges, x, costs, g, root, s_low, s_high,
max_num_iter, verbose):
""" This is the wrapper of head/tail-projection proposed in [2].
:param edges: edges in the graph.
:param x: projection vector x.
:param costs: edge costs in the graph.
:param g: the number of connected components.
:param root: root of subgraph. Usually, set to -1: no root.
:param s_low: the lower bound of the sparsity.
:param s_high: the upper bound of the sparsity.
:param max_num_iter: the maximum number of iterations used in
binary search procedure.
:param verbose: print out some information.
:return: 1. the support of the projected vector
2. the projected vector
"""
prizes = x * x
# to avoid too large upper bound problem.
if s_high >= len(prizes) - 1:
s_high = len(prizes) - 1
re_nodes = wrap_head_tail_bisearch(edges, prizes, costs, g, root, s_low,
s_high, max_num_iter, verbose)
proj_w = np.zeros_like(x)
proj_w[re_nodes[0]] = x[re_nodes[0]]
return re_nodes[0], proj_w
def algo_head_tail_bisearch(edges, x, costs, g, root, s_low, s_high,
max_num_iter, verbose):
prizes = x * x
# to avoid too large upper bound problem.
if s_high >= len(prizes) - 1:
s_high = len(prizes) - 1
re_nodes = wrap_head_tail_bisearch(edges, prizes, costs, g, root, s_low,
s_high, max_num_iter, verbose)
proj_w = np.zeros_like(x)
proj_w[re_nodes[0]] = x[re_nodes[0]]
return re_nodes[0], proj_w