def main():
n = 50000
max_dim = 45
dims = np.arange(2, max_dim + 1, step=2)
# dims = [2,5,10,15,20,25,30,35,40,45]
reps = 50
m = 0.4
p = 0.3
k = 10
page_size = 8000
num_bytes = 8
dim_percent = {d: [] for d in dims}
for dim in dims:
print('testing dimension:', dim)
M = np.floor(page_size / (2 * dim * num_bytes + num_bytes)).astype(int)
tree = RTree(M, dim, m, p)
#data = np.random.rand(n,dim)
data = np.loadtxt('../data/fv/fv' + str(dim) + 'd.txt')
data = data[:n]
#print(data.shape)
tree.insert_batch(data, np.arange(n))
for i in np.arange(reps):
#q = np.random.rand(dim)
q_idx = np.random.randint(0, n)
q = data[q_idx, :]
rtree_nns, num_visited_leaves = tree.knn_naive(q, k)
# true_nns, true_dists = knn(data, q, k)
# rtree_nns = [x.rid for x in rtree_nns]
# print('rtree nns:', rtree_nns)
# print('true nns:', true_nns)
leaves = tree.get_leaves()
total_leaves = len(leaves)
dim_percent[dim].append(num_visited_leaves / total_leaves)
print('dim:', dim, '%:', np.mean(dim_percent[dim]) * 100)
dim_percent = {k: np.mean(v) for k, v in dim_percent.items()}
percents = [dim_percent[d] for d in dim_percent.keys()]
out = np.vstack([dims, percents]).T
np.savetxt('data/rtree_real.csv', out, delimiter=',')
def main():
n = 50000
max_dim = 30
dims = np.arange(2,max_dim+1, step=2)
#dims = [2,3,4,5,10,15]
reps = 50
m = 0.4
p = 0.3
k = 10
page_size = 8000
num_bytes = 8
dim_percent = {d:[] for d in dims}
for dim in dims:
print('testing dimension:', dim)
M = np.floor(page_size/(2*dim*num_bytes + num_bytes)).astype(int)
tree = RTree(M, dim, m, p)
data = np.random.rand(n,dim)
tree.insert_batch(data, np.arange(n))
for i in np.arange(reps):
q_idx = np.random.randint(0,n)
q = data[q_idx,:]
_, num_visited_leaves = tree.knn_naive(q,k)
leaves = tree.get_leaves()
total_leaves = len(leaves)
dim_percent[dim].append(num_visited_leaves/total_leaves)
print('dim:', dim, '%:', np.mean(dim_percent[dim])*100)
dim_percent = {k:np.mean(v) for k,v in dim_percent.items()}
percents = [dim_percent[d] for d in dim_percent.keys()]
out = np.vstack([dims, percents]).T
np.savetxt('data/rtree_synthetic.csv', out, delimiter=',')
def main():
n = 50000
# 3 to 45 dimensions, step by 3
dims = np.arange(45, 48, 3)
reps = 20
m = 0.4
p = 0.3
k = 10
dim_percent = {d: [] for d in dims}
for dim in dims:
print('testing dimension:', dim)
M = np.floor(8000 / (2 * dim * 8 + 8)).astype(int)
#M = 100
tree = RTree(M, dim, m, p)
data = np.loadtxt('../data/fv/fv' + str(dim) + 'd.txt')
#data = np.random.rand(n,dim)
tree.insert_batch(data, np.arange(n))
for i in np.arange(reps):
#q = np.random.rand(dim)
# pass in the second row of dimension d as the query
rand_index = randint(0, 50000)
q = data[rand_index]
_, num_visited_leaves = tree.knn_naive(q, k)
leaves = tree.get_leaves()
total_leaves = len(leaves)
dim_percent[dim].append(num_visited_leaves / total_leaves)
dim_percent = {k: np.mean(v) for k, v in dim_percent.items()}
percents = [dim_percent[d] for d in dim_percent.keys()]
out = np.vstack([dims, percents]).T
print(out)
n = 50000
dim = 8
per_page = 8000 // (dim * 2 * 4 + 4)
# per_page = 20
print(per_page, 'entries per page')
tree = RTree(per_page, dim)
for i in range(n):
tree.insert(np.random.rand(dim), i)
if (i % (n / 10)) == 0:
print(int((i / n) * 100), '%')
print('construct time:', time.time() - start_time, 'seconds')
leaves = tree.get_leaves()
lower_mean = np.mean([l.mbb[0] for l in leaves], axis=0)
upper_mean = np.mean([l.mbb[1] for l in leaves], axis=0)
print('lower mean:', lower_mean, 'upper mean:', upper_mean)
start_time = time.time()
total_leaves = len(tree.get_leaves())
q = np.random.rand(dim)
neighbors, num_leaves = tree.knn_naive(q, 10)
print('search time:', time.time() - start_time, 'seconds')
print('leaves visited:', num_leaves, 'out of', total_leaves,
(num_leaves / total_leaves) * 100, '%')
for n in neighbors:
print(n, dist(n, q))