def main():
global args
# create the random dataset
data = generate_data(args.count, args.dim)
queries = generate_data(args.num_queries, args.dim)
# create and fill the LSH table.
t = nr_lsh.lsh_table(bits=17, dim=args.dim, num_buckets=args.count)
t.fill(data, False)
# "multiprobe" query the table.
for query in queries:
# find the nearset neighbor and its distance
nn, dist = exact_neighbor(query, data)
# the probe returns an option of KV, so it is MAYBE the
# neighbor and its id; but, it could be None.
maybe_neighbor, stats = t.probe(query, args.adj)
approx_dist1 = np.linalg.norm(maybe_neighbor[0] - query)
maybe_neighbor, stats = t.probe_approx(query, 2.0, args.adj)
if (maybe_neighbor):
approx_dist2 = np.linalg.norm(maybe_neighbor[0] - query)
else:
approx_dist2 = 0
print('{0:.2f} {1:.2f} {2:.2f}'.format(approx_dist1, approx_dist2,
dist))
def main():
global args
# create the random dataset
data = generate_data(args.count, args.dim)
queries = generate_data(args.num_queries, args.dim)
# create and fill the LSH table.
t = nr_lsh.lsh_table(num_tables=20,
bits=32,
dim=args.dim,
num_buckets=args.count)
t.fill(data, False)
# track the total time to perform all queries.
exact_total_time = 0
probe_total_time = 0
probe_approx_total_time = 0
# "multiprobe" query the table.
for query in queries:
# find the nearset neighbor and its distance
start = time.time()
nn, dist = exact_neighbor(query, data)
end = time.time()
exact_total_time += (end - start)
# probe the table and record the time.
start = time.time()
# the probe returns an option of KV, so it is MAYBE the
# neighbor and its id; but, it could be None.
maybe_neighbor, stats = t.probe(query, args.adj)
end = time.time()
probe_total_time += (end - start)
approx_dist1 = np.linalg.norm(maybe_neighbor[0] - query)
start = time.time()
maybe_neighbor, stats = t.probe_approx(query, 2.0, args.adj)
end = time.time()
probe_approx_total_time += (end - start)
if (maybe_neighbor):
approx_dist2 = np.linalg.norm(maybe_neighbor[0] - query)
else:
approx_dist2 = 0
print('{0:.2f} {1:.2f} {2:.2f}'.format(approx_dist1, approx_dist2,
dist))
print('Exact Average Time Per Query: ' +
str(exact_total_time / args.num_queries))
print('Probe Average Time Per Query: ' +
str(probe_total_time / args.num_queries))
print('Probe Approx Average Time Per Query: ' +
str(probe_approx_total_time / args.num_queries))
def main():
global args
# create the random dataset
data = generate_data(args.count, args.dim)
queries = generate_data(args.num_queries, args.dim)
# create and fill the LSH table.
mean_recalls = []
mean_times = []
for num_tables in range(5, 106, 20):
#bits, num_tables = nr_lsh.sizes_from_probs(args.count, p1, p2)
t = nr_lsh.lsh_table(num_tables=num_tables,
bits=32,
dim=args.dim,
num_buckets=args.count)
t.fill(data, False)
k_probe_recall = []
time_per_query = []
for query in queries:
# find the true topk nearset neighbor and their distances
# get indices of the true topk to compute recall.
indices, nns, dists = exact_k_neighbor(args.k, query, data)
# probe the table and record the time.
# the probe returns an option of KV, so it is MAYBE the
# neighbor and its id; but, it could be None.
start = time.time()
maybe_neighbors, stats = t.k_probe(args.k, query, args.adj)
end = time.time()
time_per_query.append(end - start)
# compute revall if no neighbors were found, recall is 0
if maybe_neighbors is None:
rec1 = 0
else:
neighbors = list(list(zip(*maybe_neighbors))[1])
rec1 = recall(neighbors, indices)
k_probe_recall.append(rec1)
mean_recalls.append(sum(k_probe_recall) / len(k_probe_recall))
mean_times.append(sum(time_per_query) / len(time_per_query))
plt.plot(mean_times, mean_recalls)
plt.show()
def main():
global args
# create the random dataset
data = generate_data(args.count, args.dim)
queries = generate_data(args.num_queries, args.dim)
# create and fill the LSH table.
t = nr_lsh.lsh_table(bits=args.bits,
dim=args.dim,
num_buckets=2**args.bits)
t.fill(data, False)
# lists to track the recalls of each type of probe.
k_probe_recall = []
k_probe_approx_recall = []
# "multiprobe" query the table.
for query in queries:
# find the true neighbors, indices in the dataset, and their distances
indices, nns, dists = exact_k_neighbor(args.k, query, data)
# the probe returns an option of KV, so it is MAYBE the
# neighbor and its id; but, it could be None.
maybe_neighbors, stats = t.k_probe(args.k, query, args.adj)
neighbors = list(list(zip(*maybe_neighbors))[1])
rec1 = recall(neighbors, indices)
k_probe_recall.append(rec1)
maybe_neighbors, stats = t.k_probe_approx(args.k, query, 0.05,
args.adj)
if (maybe_neighbors):
neighbors = list(list(zip(*maybe_neighbors))[1])
rec2 = recall(neighbors, indices)
else:
rec2 = 0
k_probe_approx_recall.append(rec2)
if args.v:
print('{0:.2f} {1:.2f}'.format(rec1, rec2))
print('mean recalls:')
print(' * k_probe: {0:.2f}'.format(
sum(k_probe_recall) / len(k_probe_recall)))
print(' * k_probe_approx: {0:.2f}'.format(
sum(k_probe_approx_recall) / len(k_probe_approx_recall)))
def main():
global args
# create the random dataset
data = generate_data(args.count, args.dim)
queries = generate_data(args.num_queries, args.dim)
# create and fill the LSH table.
t = nr_lsh.lsh_table(num_tables=50,
bits=128,
dim=args.dim,
num_buckets=args.count)
t.fill(data, False)
k_probe_recall = []
k_probe_approx_recall = []
# track the total time to perform all queries.
exact_total_time = 0
probe_total_time = 0
probe_approx_total_time = 0
# "multiprobe" query the table.
for query in queries:
# find the true topk nearset neighbor and their distances
# get indices of the true topk to compute recall.
start = time.time()
indices, nns, dists = exact_k_neighbor(args.k, query, data)
end = time.time()
exact_total_time += (end - start)
# probe the table and record the time.
start = time.time()
# the probe returns an option of KV, so it is MAYBE the
# neighbor and its id; but, it could be None.
maybe_neighbors, stats = t.k_probe(args.k, query, args.adj)
end = time.time()
probe_total_time += (end - start)
# compute revall if no neighbors were found, recall is 0
if maybe_neighbors is None:
rec1 = 0
else:
neighbors = list(list(zip(*maybe_neighbors))[1])
rec1 = recall(neighbors, indices)
k_probe_recall.append(rec1)
# find the first k items within distance c from the query
start = time.time()
maybe_neighbors, stats = t.k_probe_approx(args.k, query, 2.0, args.adj)
end = time.time()
probe_approx_total_time += (end - start)
# compute the recall. again, if no neighbors found, recall is 0.
if maybe_neighbors is None:
rec2 = 0
else:
neighbors = list(list(zip(*maybe_neighbors))[1])
rec2 = recall(neighbors, indices)
k_probe_approx_recall.append(rec2)
print('Exact Average Time Per Query: ' +
str(exact_total_time / args.num_queries))
print('Probe Average Time Per Query: ' +
str(probe_total_time / args.num_queries))
print('Probe Average Time Per Query: ' +
str(probe_approx_total_time / args.num_queries))
print('mean recalls:')
print(' * k_probe: {0:.2f}'.format(
sum(k_probe_recall) / len(k_probe_recall)))
print(' * k_probe_appeox: {0:.2f}'.format(
sum(k_probe_approx_recall) / len(k_probe_approx_recall)))