-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathelo.py
210 lines (202 loc) · 6.99 KB
/
elo.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
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
#!/usr/bin/python3.6
import numpy as np
from numpy import matlib as mb
from ROC1 import rocxy, successmatrix, unbiasedMeanMatrixVar
from warnings import filterwarnings
from DylMath import MSE, genSep
from sys import argv
from tqdm import trange, tqdm
filterwarnings('ignore')
def simulation_ELO_targetAUC(args: list, rounds: int=14, retRoc=False):
"""Args is of the form (dist, auc, n0, n1).
Rounds is how many rounds of (n0 + n1)/2 comparisons it will so.
retRoc determines if the function will return its ROC curve or its statistics.
@Author: Francesc Massanes ([email protected])
@Version: 0.1 (really beta)"""
dist, auc, n0, n1 = args
if n0 != n1:
raise NotImplementedError("n0 must equal n1")
N = n0
##
# DATA GENERATION
#
K1 = 400
K2 = 32
sep = genSep(dist, float(auc))
if dist == 'exponential':
neg = np.random.exponential(size=(N, 1))
plus = np.random.exponential(scale=sep, size=(N, 1))
elif dist == 'normal':
neg = np.random.normal(0, 1, (N, 1))
plus = np.random.normal(sep, 1, (N, 1))
else:
print("invalid argument", dist)
return None
x0 = np.array(neg)[:,0]
x1 = np.array(plus)[:,0]
empiricROC = rocxy(x1, x0)
scores = np.append(neg, plus)
truth = np.append(mb.zeros((N, 1)), mb.ones((N, 1)), axis=0)
rating = np.append(mb.zeros((N, 1)), mb.zeros((N, 1)), axis=0)
pc = list()
cnt = 0
ncmp = 0
results = list()
for eloRound in range(1, rounds+1):
toCompare = mb.zeros((2*N, 1))
if eloRound == 1:
# option A: only compare + vs -
arr = list(range(N))
#np.random.shuffle(arr)
toCompare[0::2] = np.array(arr, ndmin=2).transpose()
arr = list(range(N, 2 * N))
#np.random.shuffle(arr)
toCompare[1::2] = np.array(arr, ndmin=2).transpose()
else:
# option B: everything is valid
arr = list(range(2 * N))
np.random.shuffle(arr)
toCompare = np.array(arr, ndmin=2).transpose()
for i in range(1, 2*N, 2):
a = int(toCompare[i - 1])
b = int(toCompare[i])
QA = 10**(int(rating[a]) / K1)
QB = 10**(int(rating[b]) / K1)
EA = QA / (QA+QB)
EB = QB / (QA+QB)
if scores[a] < scores[b]:
SA = 0
SB = 1
else:
SA = 1
SB = 0
if bool(truth[a]) ^ bool(truth[b]):
if ( SA == 1 and truth[a] == 1 ):
cnt = cnt + 1
if ( SB == 1 and truth[b] == 1 ):
cnt = cnt +1
pc.append(cnt / (len(pc) + 1))
ncmp = ncmp+1
rating[a] = rating[a] + K2 * ( SA - EA )
rating[b] = rating[b] + K2 * ( SB - EB )
x0 = np.array(rating[0:N])[:,0]
x1 = np.array(rating[N:])[:,0]
roc = rocxy(x1, x0)
if retRoc:
results.append((roc, ncmp))
else:
sm = successmatrix(x1, np.transpose(x0))
auc, var = np.mean(sm), unbiasedMeanMatrixVar(sm)
mseTruth, mseEmpiric, auc = MSE(sep, dist, roc, empiricROC)
results.append((N, cnt, ncmp, var, auc, mseTruth, mseEmpiric, pc[-1]))
return results
if __name__ == '__main__':
if len(argv) > 1:
test = 1
else:
test = 3
if test == 1:
from main import multiRunner
multiRunner(simulation_ELO_targetAUC, "Elo")
elif test == 2:
animation = False
if animation:
import matplotlib.pyplot as plt
from matplotlib.animation import FuncAnimation
from matplotlib.animation import PillowWriter
frames = 100
results = simulation_ELO_targetAUC(('normal', 0.8853, 128, 128), rounds=frames)
fig, ax = plt.subplots()
fig.set_tight_layout(True)
pbar = tqdm(total=len(results))
def update(i):
"""Update the frame"""
pbar.update()
label = f"timestep {i}"
ax.clear()
roc = rocxy(*results[i])
ax.plot(roc['x'], roc['y'])
ax.set_title(f"{i:02d}")
return label, ax
anim = FuncAnimation(fig, update, frames=np.arange(0, frames), interval=100)
anim.save("rocs.gif", writer=PillowWriter(fps=10))
pbar.close()
else:
import matplotlib.pyplot as plt
from apng import APNG
from DylSort import treeMergeSort
from DylComp import Comparator
from DylData import continuousScale
from DylMath import genROC, avROC
seed = 15
data, D0, D1 = continuousScale(128, 128)
comp = Comparator(data, level=0, rand=True, seed=seed)
comp.genRand(len(D0), len(D1), 7.72, 'exponential')
np.random.seed(seed)
im = APNG()
fig, (ax1, ax2) = plt.subplots(nrows=1, ncols=2)
fig.suptitle("Pass number, MSE true, MSE empirical")
x = np.linspace(0, 1, num=200)
y = x**(1/7.72)
ax1.set_aspect('equal', 'box')
ax2.set_aspect('equal', 'box')
elo = simulation_ELO_targetAUC(True)
merge = treeMergeSort(data, comp, statParams=[(D0, D1)], combGroups=False)
plt.tight_layout()
for i in trange(8):
roc, mseTheo, mseEmp, empiricROC = next(elo)
ax1.plot(x, y, linestyle='--', label='true', lw=3)
ax1.plot(empiricROC['x'], empiricROC['y'], linestyle=':', lw=2, label='empirical')
ax1.plot(roc['x'], roc['y'], label='predicted')
ax1.legend(loc=4)
ax1.set_title(f"ELO\n{i+1}, {mseTheo[0]*1000:02.3f}E(-3), {mseEmp[0]*1000:02.3f}E(-3)")
groups = next(merge)
rocs = []
for group in groups:
rocs.append(genROC(group, D0, D1))
roc = avROC(rocs)
mseTheo, mseEmp, auc = MSE(7.72, 'exponential', zip(*roc)), MSE(7.72, 'exponential', zip(*roc), zip(empiricROC['x'], empiricROC['y']))
ax2.plot(x, y, linestyle='--', label='true', lw=3)
ax2.plot(empiricROC['x'], empiricROC['y'], linestyle=':', lw=2, label='empirical')
ax2.plot(*roc, label='predicted')
ax2.legend()
ax2.set_title(f"merge\n{i+1}, {mseTheo[0]*1000:02.3f}E(-3), {mseEmp[0]*1000:02.3f}E(-3)")
plt.savefig("both")
im.append_file("both.png", delay=1000)
ax1.clear()
ax2.clear()
im.save("both.png")
elif test == 3:
from DylComp import Comparator
from DylSort import treeMergeSort
from DylData import continuousScale
from DylMath import genROC, avROC
import matplotlib.pyplot as plt
data, D0, D1 = continuousScale(128, 128)
comp = Comparator(data, rand=True, level=0, seed=20)
results = [res for res in simulation_ELO_targetAUC(('normal', 0.8853, 128, 128), retRoc=True)]
mergeResults = list()
for groups in treeMergeSort(data, comp, combGroups=False):
rocs = list()
for group in groups:
roc = genROC(group, D1, D0)
rocs.append(roc)
mergeResults.append((avROC(rocs), len(comp)))
matches = [(np.argmin([abs(res[1] - mergeLen) for res in results])) for (groups, mergeLen) in mergeResults]
fig, axes = plt.subplots(nrows=4, ncols=2, sharex=True, sharey=True)
for axn, layer in enumerate([0, 3, 5, 7]):
axes[axn][0].plot(mergeResults[layer][0][1], mergeResults[layer][0][0], 'r')
axes[axn][1].plot(results[matches[layer]][0]['x'], results[matches[layer]][0]['y'], 'r')
axes[axn][0].plot(comp.empiricROC()['x'], comp.empiricROC()['y'], 'b:')
axes[axn][1].plot(comp.empiricROC()['x'], comp.empiricROC()['y'], 'b:')
axes[axn][0].set_aspect('equal', 'box')
axes[axn][1].set_aspect('equal', 'box')
axes[axn][0].set_title(str(mergeResults[layer][1]))
axes[axn][1].set_title(str(results[matches[layer]][1]))
fig.suptitle("Merge Massanes")
figManager = plt.get_current_fig_manager()
figManager.window.showMaximized()
fig.subplots_adjust(hspace=0.24)
for i in range(10):
fig.tight_layout()
plt.show()