def make_graphs(n_tmplt_nodes, n_world_nodes, n_layers, tmplt_prob, world_prob):
tmplt_nodes = [x for x in range(n_tmplt_nodes)]
world_nodes = [x for x in range(n_world_nodes)]
tmplt_shape = (n_tmplt_nodes, n_tmplt_nodes)
world_shape = (n_world_nodes, n_world_nodes)
tmplt_adj_mats = [csr_matrix(np.random.choice([0, 1], tmplt_shape, p=[1-tmplt_prob, tmplt_prob])) for i in range(n_layers)]
world_adj_mats = [csr_matrix(np.random.choice([0, 1], world_shape, p=[1-world_prob, world_prob])) for i in range(n_layers)]
channels = [str(x) for x in range(n_layers)]
tmplt = uclasm.Graph(np.array(tmplt_nodes), channels, tmplt_adj_mats)
world = uclasm.Graph(np.array(world_nodes), channels, world_adj_mats)
tmplt.is_cand = np.ones((tmplt.n_nodes,world.n_nodes), dtype=np.bool)
tmplt.candidate_sets = {x: set(world.nodes) for x in tmplt.nodes}
return tmplt, world
import sys
sys.path.append("..")
import numpy as np
import uclasm
nodelist, channels, adjs = uclasm.load_combo(
"example_data_files/example_combo.csv",
node_vs_edge_col=0,
node_str="v",
src_col=1,
dst_col=2,
channel_col=3,
node_col=1,
label_col=2,
header=0)
nodes = nodelist.node
labels = nodelist.label
# Use the same graph data for both template and world graphs
tmplt = uclasm.Graph(nodes, channels, adjs, labels=labels)
world = uclasm.Graph(nodes, channels, adjs, labels=labels)
link_col_square[j * 3:j * 3 + 3, size + i * 3 + j] = 1
for i in range(size):
digit_idxs = [i * size + j for j in range(size)]
rs_idxs = digit_idxs + [x + 2 * size * size for x in digit_idxs]
cs_idxs = [x + size * size for x in digit_idxs
] + [x + 2 * size * size for x in digit_idxs]
world_adj_mats[1][np.ix_(rs_idxs, rs_idxs)] = link_row_square
world_adj_mats[1][np.ix_(cs_idxs, cs_idxs)] = link_col_square
for i in range(2):
world_adj_mats[i] = sp.sparse.csr_matrix(world_adj_mats[i])
tmplt_adj_mats[i] = sp.sparse.csr_matrix(tmplt_adj_mats[i])
# initial candidate set for template nodes is the full set of world nodes
tmplt = uclasm.Graph(np.array(tmplt_nodes), channels, tmplt_adj_mats)
world = uclasm.Graph(np.array(world_nodes), channels, world_adj_mats)
tmplt.is_cand = np.ones((tmplt.n_nodes, world.n_nodes), dtype=np.bool)
tmplt.candidate_sets = {x: set(world.nodes) for x in tmplt.nodes}
def update_node_candidates(tmplt, world, tmplt_node, cands):
cand_row = np.zeros(world.n_nodes, dtype=np.bool)
for cand in cands:
cand_row[world.node_idxs[cand]] = True
tmplt.is_cand[tmplt.node_idxs[tmplt_node]] &= cand_row
tmplt.labels = np.array(['__'] * tmplt.n_nodes)
world.labels = np.array(['__'] * world.n_nodes)
if stype == "9x9":
# Now generate world and template graphs, inserting a signal in the top left
np.random.seed(0)
channels = list(range(n_channels))
# TODO: allow for nodes of arbitrary dtype
# Note: by design, the template nodes have the same ids as the signal nodes
tmplt_nodes = np.arange(n_world_nodes, n_world_nodes+n_tmplt_nodes)
world_nodes = np.arange(n_world_nodes, 2*n_world_nodes)
tmplt_adj_mats = []
world_adj_mats = []
for channel in channels:
tmplt_adj = np.random.geometric(tmplt_p, (n_tmplt_nodes, n_tmplt_nodes)) - 1
world_adj = np.random.geometric(world_p, (n_world_nodes, n_world_nodes)) - 1
# Embed a signal in the top left block of the world graph
world_adj[:n_tmplt_nodes, :n_tmplt_nodes] += tmplt_adj
tmplt_adj_mats.append(sparse.csc_matrix(tmplt_adj))
world_adj_mats.append(sparse.csc_matrix(world_adj))
# initial candidate set for template nodes is the full set of world nodes
tmplt = uclasm.Graph(tmplt_nodes, channels, tmplt_adj_mats)
world = uclasm.Graph(world_nodes, channels, world_adj_mats)