-
Notifications
You must be signed in to change notification settings - Fork 1
/
compute_distance_metrics.py
166 lines (136 loc) · 5.4 KB
/
compute_distance_metrics.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
import numpy as np
from scipy.stats import pearsonr, kendalltau
from hashfunctions import *
import xxhash
mx_hash = 4.5*10**9
mn_hash = 5*10**6
T = np.transpose
def novelty_dist(S,q,dist_met):
if dist_met == 'cos':
cs = -(np.max([cos_sim(s,q) for s in S])-1)
elif dist_met == 'cor':
cs = -(np.max([cor_sim(s,q) for s in S])-1)
elif dist_met == 'euc':
cs = np.min([euc_sim(s,q) for s in S])
return cs
def closest_ind(S,q): # to test ORN output
return np.argmax([cos_sim(s,q) for s in S])
def cos_sim(a,b):
return np.dot(a,b)/(np.sqrt(np.sum(a**2)) * np.sqrt(np.sum(b**2)))
def euc_sim(a,b):
return np.sqrt(np.sum((a-b)**2))
def tag_cnt(S,q):
return np.max([np.sum(s[q==1]) for s in S])
# hash function
def get_hash(v,m,k):
# using v, get k hashes in range [0,m]
val = np.sum(v)
this_hash = (np.array([xxhash.xxh32(str(val+i)).intdigest() for i in range(k)])-mn_hash)/(mx_hash-mn_hash)
return this_hash*m
def hash_array(A,m,k):
shp = np.shape(A)
if len(shp)==2:
B = list()
for i in range(shp[0]):
B.append(get_hash(A[i],m,k))
elif len(shp)==3: # recursive approach
B = list()
for i in range(shp[0]):
B.append(hash_array(A[i],m,k))
return np.array(B).astype(int)
# utility functions
def min_m(n,eps):
return int(np.ceil(n*np.log2(np.e)*np.log2(1/eps)))
def opt_k(m,n):
return int(np.floor(np.log(2)*m/n))
def e_eps(m,k,n):
# From Broder&Mitenmacher2004
p = (1.0-1.0/m)**(k*n)
return (1-p)**k
def cor_fun(x1,x2):
a1 = pearsonr(x1,x2)[0]
a2 = kendalltau(x1,x2)[0]
return np.round((a1,a2),2)
def get_distance_metrics(S,q,m=None,k=None,eps=None,dist_met = 'both',proj='SB4',app_str='',text_out=True,
ORN_SPECIAL=False,VERBOSE=0):
# must specify either m and k or eps (FP rate)
n_ex,l_ex,dIn = np.shape(S)
if not m:
m = min_m(l_ex,eps)
k = opt_k(m,l_ex)
if not eps:
eps = e_eps(m,l_ex,k)
if proj == 'DG':
M = np.random.randn(dIn,m)
elif proj[:2]=='SB':
M = T(create_projection_matrix(dIn,m,proj))
else:
assert(0)
Sm = np.matmul(S,M) # n_ex, l_ex, m
qm = np.matmul(q,M) # n_ex, m
Sm_LSH = np.argpartition(Sm,-k)[:,:,-k:] # grab the k largest values
qm_LSH = np.argpartition(qm,-k)[:,-k:]
Sm_LSH_tag = np.zeros(np.shape(Sm))
for i,ii in np.ndenumerate(Sm_LSH): # produce Boolean tags
Sm_LSH_tag[i[0],i[1],ii] = 1
qm_LSH_tag = np.zeros(np.shape(qm))
for i,ii in enumerate(qm_LSH):
qm_LSH_tag[i,ii] = 1
# bloom filter
Sm_LSH_bloom = np.sum(Sm_LSH_tag,axis=1)>0
#avg_hash_loc = np.mean(Sm_LSH_bloom,axis=0)
N_LSH_tag = [-(tag_cnt(Sm_LSH_tag[i],qm_LSH_tag[i])/k-1) for i in range(n_ex)] # Normalized
N_LSH_bloom = -(np.sum([Sm_LSH_bloom[i,qm_LSH[i]] for i in range(n_ex)],axis=1)/k-1)
Sm_hash = hash_array(S,m,k)
qm_hash = hash_array(q,m,k)
Sm_hash_tag = np.zeros(np.shape(Sm))
for i,ii in np.ndenumerate(Sm_hash): # produce Boolean tags
Sm_hash_tag[i[0],i[1],ii] = 1
qm_hash_tag = np.zeros(np.shape(qm))
for i,ii in enumerate(qm_hash):
qm_hash_tag[i,ii] = 1
# non-local bloom filter
Sm_hash_bloom = (np.sum(Sm_hash_tag,axis=1)>0)*1 # *1 to make an integer, not a boolean
#avg_hash_loc = np.mean(Sm_hash_bloom,axis=0)
# closest hash
N_hash_tag = [-(tag_cnt(Sm_hash_tag[i],qm_hash_tag[i])/k-1) for i in range(n_ex)] # Normalized
N_hash_bloom = -(np.sum([Sm_hash_bloom[i,qm_hash[i]] for i in range(n_ex)],axis=1)/k-1)
# get metrics to print
eps_str = '%.1e' %eps
nam_str = (app_str + '_d1:' + str(dIn) + '_PROJ:' + proj + '_' +
dist_met + '_m:'+ str(m) +'_k:'+str(k)+'_eps:'+eps_str+' ' )
if dist_met == 'both':
N_cos = [novelty_dist(S[i],q[i],'cos') for i in range(n_ex)]
N_euc = [novelty_dist(S[i],q[i],'euc') for i in range(n_ex)]
m_0c = 'OGDIST:%.2f' %np.mean(N_cos) # dist ORN space
m_1c = cor_fun(N_cos,N_LSH_tag)
m_2c = cor_fun(N_cos,N_LSH_bloom)
m_3c = cor_fun(N_cos,N_hash_bloom)
m_0e = 'OGDIST:%.2f' %np.mean(N_euc) # dist ORN space
m_1e = cor_fun(N_euc,N_LSH_tag)
m_2e = cor_fun(N_euc,N_LSH_bloom)
m_3e = cor_fun(N_euc,N_hash_bloom)
else:
N = [novelty_dist(S[i],q[i],dist_met) for i in range(n_ex)]
m_0c = 'OGDIST:%.2f' %np.mean(N) # dist ORN space
m_1c = cor_fun(N,N_LSH_tag)
m_2c = cor_fun(N,N_LSH_bloom)
m_3c = cor_fun(N,N_hash_bloom)
## get quick ORN special [grab index that was closest in ORN space]
if ORN_SPECIAL:
ind_closest_ORN = [closest_ind(S[i],q[i]) for i in range(n_ex)]
ORN_SP_OUT = [-( np.sum(Sm_LSH_tag[i,ind_closest_ORN[i],qm_LSH[i]]) )/k-1 for i in range(n_ex)]
m_3c = cor_fun(N_cos,ORN_SP_OUT)
m_3e = cor_fun(N_euc,ORN_SP_OUT)
mPRT = str([m_1c,m_2c,m_3c])
mPRT = ' PROJ: %.2f %.2f, LSHBLOOM: %.2f %.2f, HBLOOM %.2f %.2f' %tuple(np.concatenate([m_1c, m_2c, m_3c]))
if VERBOSE==1:
print(nam_str + m_0c+ mPRT)
elif VERBOSE==2:
print(".",end=" ",flush=True)
if text_out:
return (nam_str + m_0c+ mPRT)
elif dist_met == 'both':
return(m_1c[0],m_1e[0],m_2c[0],m_2e[0],m_3c[0],m_3e[0])
else:
return (m_1c[0],m_2c[0],m_3c[0])