def _break_big_blocks(self, dist_maker: DistanceMetric) -> None:
block_copy = self.blocks.copy()
for block_id, block_members in enumerate(block_copy):
if len(block_members) <= (2 * (self.k + 1)):
continue
distances = dist_maker.pairwise(self.Xt[block_members, :])
max_dist = np.max(distances)
loc_i, loc_j = np.where(distances == max_dist)
seed_i = block_members[loc_i[0]]
seed_j = block_members[loc_j[0]]
members_i = self.knn[seed_i]
members_j = [k for k in self.knn[seed_j] if k not in members_i]
loners = [
ind for ind, k in enumerate(block_members)
if k not in members_i and k not in members_j
]
loner_i_dist = distances[loners, loc_i[0]].reshape(-1)
loner_j_dist = distances[loners, loc_j[0]].reshape(-1)
for i_loner, loner in enumerate(loners):
if loner_i_dist[i_loner] <= loner_j_dist[i_loner]:
members_i = np.append(members_i, block_members[loner])
else:
members_j = np.append(members_j, block_members[loner])
if len(members_j) == 1:
neighbor = self.knn[seed_j][1]
members_j = np.array([seed_j, neighbor])
members_i = [m for m in members_i if m != neighbor]
self.blocks[block_id] = members_i
for member in members_i:
self.block_membership[member] = block_id
self.blocks.append(members_j)
for member in members_j:
self.block_membership[member] = len(self.blocks) - 1
def cdf_comparison(rerouting_experiment):
# Load pre-processed data for path control
data3, paco_hist3, rero_hist3, pc3 = load_all_runs(rerouting_experiment)
# Load geo information
altitude = 1300
min_elev = 40
orbits = 32
sat_per_orbit = 50
inclination = 53
GST_FILE = "data/raw/ixp_geolocation.csv"
SRC_FILE = "data/raw/WUP2018-F22-Cities_Over_300K_Annual.csv"
sat_pos, gst_pos, src_pos = load_locations(altitude, orbits, sat_per_orbit,
inclination, GST_FILE, SRC_FILE)
# Distance between cities
pairwise_src = DistanceMetric.pairwise(
DistanceMetric.get_metric("haversine"), src_pos, src_pos)
pairwise_src = haversine_to_km(pairwise_src)
terr_only = pairwise_src * FIBER_PATH_STRETCH / LIGHT_IN_FIBER
gclatency = pairwise_src / LIGHT_IN_FIBER
gcbound = pairwise_src / LIGHT_IN_VACUUM
gc_hist, buckets = np.histogram(gclatency * 1000, bins=100, range=(0, 120))
gcb_hist, _ = np.histogram(gcbound * 1000, bins=100, range=(0, 120))
terr_hist, _ = np.histogram(terr_only * 1000, bins=100, range=(0, 120))
plt.figure(figsize=(8, 6))
plt.plot(buckets[1:],
np.cumsum(gc_hist) / np.sum(gcb_hist),
":",
markersize=4,
markevery=[50],
label="Great-circle fiber-speed",
linewidth=3)
plt.plot(buckets[1:],
np.cumsum(gcb_hist) / np.sum(gcb_hist),
':',
markersize=4,
markevery=[40],
label="Great-circle c-speed",
linewidth=3)
plt.plot(buckets[1:],
np.cumsum(terr_hist) / np.sum(gcb_hist),
':',
markersize=4,
markevery=[30],
label="Great-circle path",
linewidth=3)
plt.xlabel("One-way latency (ms)")
plt.ylabel("CDF")
x_axis = data3[10]['paco']['hist'][0][1][:-1] * 1000
paco_avg = np.asarray(paco_hist3)
paco_avg = np.average(paco_avg, axis=0)
rero_avg = np.asarray(rero_hist3)
rero_avg = np.average(rero_avg, axis=0)
plt.plot(x_axis,
np.cumsum(paco_avg) / np.sum(paco_avg),
'-',
markersize=4,
markevery=[40],
label="PaCo avg.",
linewidth=3)
plt.plot(x_axis,
np.cumsum(rero_avg) / np.sum(rero_avg),
'-.',
markersize=4,
markevery=[30],
label="ReRo avg.",
linewidth=3)
plt.legend()
plt.xlim(0, 120)
plt.ylim(0, 1)
# Save figure
plt.savefig("figures/paco-rero-cdf.png")
def _shortest_dist_between_cluster(X: DataFrame, dist: DistanceMetric,
labels: np.ndarray, label_i: int,
label_j: int) -> float:
return np.min(
dist.pairwise(X[labels == label_i], X[labels == label_j]))
def _cluster_diameter(X: DataFrame, dist: DistanceMetric,
labels: np.ndarray, label: int) -> float:
return np.max(dist.pairwise(X[labels == label]))
def _sum_distance_in_cluster(X: DataFrame, dist: DistanceMetric,
labels: np.ndarray, label: int) -> float:
return np.sum(dist.pairwise(X[labels == label]))
