/
nn_algorithms.py
182 lines (146 loc) · 6.24 KB
/
nn_algorithms.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
import numpy as np
from nearpy import Engine
from nearpy.hashes import RandomBinaryProjections
from nearpy.hashes import PCABinaryProjections
from sklearn.neighbors import NearestNeighbors
from rbtree import RBTree
class LSHNeighbors(NearestNeighbors):
def __init__(self, n_neighbors=5,
radius=1.0,
algorithm='auto',
leaf_size=30,
metric='minkowski',
p=2,
metric_params=None,
n_jobs=None,
**kwargs):
super(LSHNeighbors, self).__init__(n_neighbors=n_neighbors,
radius=radius,
algorithm=algorithm,
leaf_size=leaf_size,
metric=metric,
p=p,
metric_params=metric_params,
n_jobs=n_jobs,
**kwargs)
def fit(self, X, y=None, hash="randbinary"):
X = np.array(X)
assert len(X.shape) == 2, "X not 2-rank"
dimension = X.shape[-1]
if hash == "randbinary":
rbp = RandomBinaryProjections('rbp', 10)
elif hash == "pcabinary":
rbp = PCABinaryProjections('rbp', 10, training_set=X)
self.engine = Engine(dimension, lshashes=[rbp])
index = 0
for x in X:
self.engine.store_vector(x, str(index))
index += 1
# count += index
def kneighbors(self, X=None, n_neighbors=None, return_distance=True):
if len(X.shape) == 1:
results = self.engine.neighbours(X)
# dists = [elem[2] for elem in results]
if n_neighbors == None:
n_neighbors = len(results)
else:
n_neighbors = min(len(results), n_neighbors)
dists = np.array([np.linalg.norm(X - elem[0]) for elem in results])
indices = np.array([int(elem[1]) for elem in results])
vectors = np.array([elem[0] for elem in results])
return (dists[:n_neighbors],
indices[:n_neighbors],
vectors[:n_neighbors])
elif len(X.shape) == 2:
results = [self.engine.neighbours(x) for x in X]
if n_neighbors == None:
n_neighbors = min([len(result) for result in results])
dists = np.array([[np.linalg.norm(X - result[i][0]) for i in range(min(len(result), n_neighbors))] for result in results])
indices = np.array([[int(result[i][1]) for i in range(min(len(result), n_neighbors))] for result in results])
vectors = np.array([[result[i][0] for i in range(min(len(result), n_neighbors))] for result in results])
return (dists, indices, vectors)
else:
raise ValueError('X has rank higher than 2')
def kneighbors_graph(self, X=None, n_neighbors=None, mode='connectivity'):
# Almost definitely don't need this
raise NotImplementedError
def radius_neighbors(self, X=None, radius=None, return_distance=True):
raise NotImplementedError
def radius_neighbors_graph(self, X=None, radius=None, mode='connectivity'):
# Don't need this either
raise NotImplementedError
def set_params(self, **params):
raise NotImplementedError
class DCINeighbors(NearestNeighbors):
"""
Dynamic Continuous Indexing: https://arxiv.org/pdf/1512.00442.pdf
"""
def __init__(self, n_neighbors=5,
radius=1.0,
algorithm='auto',
leaf_size=30,
metric='minkowski',
p=2,
metric_params=None,
n_jobs=None,
**kwargs):
super(DCINeighbors, self).__init__(n_neighbors=n_neighbors,
radius=radius,
algorithm=algorithm,
leaf_size=leaf_size,
metric=metric,
p=p,
metric_params=metric_params,
n_jobs=n_jobs,
**kwargs)
def fit(self, X, m=25, L=2, y=None):
assert len(X.shape) ==2, "X must be 2-rank"
n = X.shape[0]
d = X.shape[1]
U = np.random.normal(size=(d, m * L))
U = U / np.linalg.norm(U, axis=0, keepdims=True)
T = [RBTree() for _ in range(m * L)]
projected_data = np.matmul(X, U) #all the pbars
for j in range(m):
for l in range(L):
for i in range(n):
T[j + m * l].insert(KeyValeWrap(P[i][j + m * l], i))
self.T = T
self.X = X
self.d = d
self.n = n
self.U = U
self.L = L
self.m = m
self.P = projected_data
raise NotImplementedError
def kneighbors(self, X=None, n_neighbors=None, eps=1.0, return_distance=True):
U = self.U
C = np.zeros(shape=[self.L, self.n])
qbar = np.matmul(U.T, X)
S = [None for _ in range(self.L)]
# for i in range(self.n):
# for l in range(self.L):
# for j in range(self.m):
def kneighbors_graph(self, X=None, n_neighbors=None, mode='connectivity'):
# Almost definitely don't need this
raise NotImplementedError
def radius_neighbors(self, X=None, radius=None, return_distance=True):
raise NotImplementedError
def radius_neighbors_graph(self, X=None, radius=None, mode='connectivity'):
# Don't need this either
raise NotImplementedError
def set_params(self, **params):
raise NotImplementedError
class KeyValWrap:
def __init__(self, key, val):
self.key = key
self.val = val
def __lt__(self,other):
return (self.key<other.key)
def __le__(self,other):
return (self.key<=other.key)
def __gt__(self,other):
return (self.key>other.key)
def __ge__(self,other):
return (self.key>=other.key)