def build_index(self, X):
f = X.shape[1]
n = X.shape[0]
lsh_forest = LSH_forest(number_of_trees=10)
lsh_forest.build_index(X)
return lsh_forest
def build_index(self, X):
f = X.shape[1]
n = X.shape[0]
lsh_forest = LSH_forest(number_of_trees=10)
lsh_forest.build_index(X)
return lsh_forest
def precision_test_candidates_LSH_F_random(X, limits = [10, 100, 1000], iterations = 100):
f = X.shape[1]
n = X.shape[0]
t = LSH_forest(number_of_trees=10)
t.build_index(X)
limits = limits
k = 10
prec_sum = {}
prec_n = iterations
time_sum = {}
candidate_sum = {}
for i in xrange(prec_n):
query_vector = np.random.random(size=f)
print 'finding nbs for a random vector'
neighbors, candidates = t.query_num_candidates(query_vector, c=n)
closest = set(neighbors[:k])
for limit in limits:
t0 = time.time()
neighbors, candidates= t.query_num_candidates(query_vector, c=limit)
T = time.time() - t0
toplist = neighbors[:k]
found = len(closest.intersection(toplist))
hitrate = 1.0 * found / k
prec_sum[limit] = prec_sum.get(limit, 0.0) + hitrate
time_sum[limit] = time_sum.get(limit, 0.0) + T
candidate_sum[limit] = candidate_sum.get(limit, 0) + candidates
for limit in limits:
print 'limit: %.8f precision: %6.2f%% avg time: %.6fs avg_candiates: %.1f' % (limit, 100.0 *
prec_sum[limit] / (i + 1), time_sum[limit] / (i + 1),
candidate_sum[limit]/ (i+1))
return prec_sum, time_sum, candidate_sum
def precision_test_candidates_LSH_F_random(X,
limits=[10, 100, 1000],
iterations=100):
f = X.shape[1]
n = X.shape[0]
t = LSH_forest(number_of_trees=10)
t.build_index(X)
limits = limits
k = 10
prec_sum = {}
prec_n = iterations
time_sum = {}
candidate_sum = {}
for i in xrange(prec_n):
query_vector = np.random.random(size=f)
print 'finding nbs for a random vector'
neighbors, candidates = t.query_num_candidates(query_vector, c=n)
closest = set(neighbors[:k])
for limit in limits:
t0 = time.time()
neighbors, candidates = t.query_num_candidates(query_vector,
c=limit)
T = time.time() - t0
toplist = neighbors[:k]
found = len(closest.intersection(toplist))
hitrate = 1.0 * found / k
prec_sum[limit] = prec_sum.get(limit, 0.0) + hitrate
time_sum[limit] = time_sum.get(limit, 0.0) + T
candidate_sum[limit] = candidate_sum.get(limit, 0) + candidates
for limit in limits:
print 'limit: %.8f precision: %6.2f%% avg time: %.6fs avg_candiates: %.1f' % (
limit, 100.0 * prec_sum[limit] / (i + 1), time_sum[limit] /
(i + 1), candidate_sum[limit] / (i + 1))
return prec_sum, time_sum, candidate_sum
import numpy as np
from lsh_forest import LSH_forest
import matplotlib.pyplot as plt
from sklearn.metrics import euclidean_distances
"""
Create a dummy 2 dimensional data set for the visualization.
Create LSH forest with a single tree and build index with the dummy data.
"""
samples = 10000
dummy_x = np.random.rand(samples,2)
lshf = LSH_forest(number_of_trees=1)
lshf.build_index(dummy_x)
#Get candidate neighbors for a query
point = dummy_x[np.random.randint(0,samples)]
#point = np.random.rand(1,2)[0] #Use this if a random vector is required
neighbors, candidates = lshf.query_candidates(point, m=20)
#Plot candidate distribution with the query
x = dummy_x[[candidates],0]
y = dummy_x[[candidates],1]
plt.scatter(x, y, s=10, c='g')
plt.scatter(point[0], point[1], s=20, c='r')
plt.ylabel('Y')
plt.xlabel('X')
plt.title("Candidates distribution")
plt.show()
