def pTestGivenDis(d):
if d == "болен":
return DDist({"положительный": 0.99, "отрицательный": 0.01})
elif d == "здоров":
return DDist({"положительный": 0.001, "отрицательный": 0.999})
else:
raise Exception("неправильное значение для D")
def _oodelayPoisson(ddist, lambd):
cmf = []
tcmf = []
#front dist
for i in range(len(ddist.cmfy)):
s = 0
#front part
for j in range(i, len(ddist.cmfy)):
s += (1 - ddist.cmfy[j]) * ddist.h
#tail part
for j in range(len(ddist.tcmfy)):
s += (1 - ddist.tcmfy[j]) * ddist.th
p = ddist.cmfy[i] * np.exp(-lambd * s)
cmf.append(p)
#tail dist
for i in range(len(ddist.tcmfy)):
s = 0
for j in range(i, len(ddist.tcmfy)):
s += (1 - ddist.tcmfy[j]) * ddist.th
p = ddist.tcmfy[i] * np.exp(-lambd * s)
tcmf.append(p)
pmf = [cmf[0]]
for i in range(1, len(cmf)):
pmf.append(cmf[i] - cmf[i - 1])
if len(tcmf) == 0:
return DDist(ddist.lb, pmf, ddist.h, [], ddist.th)
else:
tpmf = [tcmf[0] - cmf[-1]]
for i in range(1, len(tcmf)):
tpmf.append(tcmf[i] - tcmf[i - 1])
return DDist(ddist.lb, pmf, ddist.h, tpmf, ddist.th)
def _oodelayFixedInterval(ddist, intvl):
cmf = []
tcmf = []
#front dist
for i in range(len(ddist.cmfy)):
curr = 1
prev = 0
#front part
start = ddist.lb + i * ddist.h
x = start
while x < ddist.tb:
j = int(floor((x - ddist.lb) / ddist.h))
prev = curr
curr *= ddist.cmfy[j]
x += intvl
if curr < 1e-10:
x = ddist.ub
break
if abs(prev - curr) / curr < 1e-6:
x = ddist.ub
break
#tail part
while x < ddist.ub:
j = int(floor((x - ddist.tb) / ddist.th))
prev = curr
curr *= ddist.tcmfy[j]
x += intvl
if curr < 1e-10:
break
if abs(prev - curr) / curr < 1e-6:
break
cmf.append(curr)
#tail dist
for i in range(len(ddist.tcmfy)):
curr = 1
prev = 0
start = ddist.tb + i * ddist.th
x = start
while x < ddist.ub:
j = int(floor((x - ddist.tb) / ddist.th))
prev = curr
curr *= ddist.tcmfy[j]
x += intvl
if curr < 1e-10:
break
if abs(prev - curr) / curr < 1e-6:
break
tcmf.append(curr)
pmf = [cmf[0]]
for i in range(1, len(cmf)):
pmf.append(cmf[i] - cmf[i - 1])
if len(tcmf) == 0:
return DDist(ddist.lb, pmf, ddist.h, [], ddist.th)
else:
tpmf = [tcmf[0] - cmf[-1]]
for i in range(1, len(tcmf)):
tpmf.append(tcmf[i] - tcmf[i - 1])
return DDist(ddist.lb, pmf, ddist.h, tpmf, ddist.th)
def testMaxn():
print '===== test maxn ====='
print '\n>>>normal\n'
mu = 10
sigma = 3
n = 5
x = []
y = []
for i in range(100000):
m = -sys.maxint
for j in range(n):
r = np.random.normal(mu, sigma)
x.append(r)
if r > m:
m = r
y.append(m)
ddist0 = DDist.create(x)
ddist1 = DDist.create(y)
ddist2 = maxn(n, ddist0)
print 'base', ddist0.mean, ddist0.std
print 'sim', ddist1.mean, ddist1.std
print 'ana', ddist2.mean, ddist2.std
ddist1.plot('/tmp/test_maxn_normal1.pdf')
ddist2.plot('/tmp/test_maxn_normal2.pdf')
print '\n>>>pareto\n'
a = 1.3
x0 = 0
x = []
y = []
for i in range(100000):
m = -sys.maxint
for j in range(n):
r = np.random.pareto(a) + x0
x.append(r)
if r > m:
m = r
y.append(m)
ddist0 = DDist.create(x, tail=('p', 0.1), tnh=25)
ddist1 = DDist.create(y, tail=('p', 0.1), tnh=25)
#ddist0 = DDist.create(x)
#ddist1 = DDist.create(y)
ddist2 = maxn(n, ddist0)
print 'base', ddist0.mean, ddist0.std, 'length', len(ddist0.pmfy), len(
ddist0.tpmfy)
print 'sim', ddist1.mean, ddist1.std, 'length', len(ddist1.pmfy), len(
ddist1.tpmfy)
print 'ana', ddist2.mean, ddist2.std, 'length', len(ddist2.pmfy), len(
ddist2.tpmfy)
ddist1.plot('/tmp/test_maxn_pareto1.pdf')
ddist2.plot('/tmp/test_maxn_pareto2.pdf')
print '===== end =====\n'
def testMaxn():
print '===== test maxn ====='
print '\n>>>normal\n'
mu = 10
sigma = 3
n = 5
x = []
y = []
for i in range(100000):
m = -sys.maxint
for j in range(n):
r = np.random.normal(mu, sigma)
x.append(r)
if r > m:
m = r
y.append(m)
ddist0 = DDist.create(x)
ddist1 = DDist.create(y)
ddist2 = maxn(n, ddist0)
print 'base', ddist0.mean, ddist0.std
print 'sim', ddist1.mean, ddist1.std
print 'ana', ddist2.mean, ddist2.std
ddist1.plot('/tmp/test_maxn_normal1.pdf')
ddist2.plot('/tmp/test_maxn_normal2.pdf')
print '\n>>>pareto\n'
a = 1.3
x0 = 0
x = []
y = []
for i in range(100000):
m = -sys.maxint
for j in range(n):
r = np.random.pareto(a) + x0
x.append(r)
if r > m:
m = r
y.append(m)
ddist0 = DDist.create(x, tail=('p', 0.1), tnh=25)
ddist1 = DDist.create(y, tail=('p', 0.1), tnh=25)
#ddist0 = DDist.create(x)
#ddist1 = DDist.create(y)
ddist2 = maxn(n, ddist0)
print 'base', ddist0.mean, ddist0.std, 'length', len(ddist0.pmfy), len(ddist0.tpmfy)
print 'sim', ddist1.mean, ddist1.std, 'length', len(ddist1.pmfy), len(ddist1.tpmfy)
print 'ana', ddist2.mean, ddist2.std, 'length', len(ddist2.pmfy), len(ddist2.tpmfy)
ddist1.plot('/tmp/test_maxn_pareto1.pdf')
ddist2.plot('/tmp/test_maxn_pareto2.pdf')
print '===== end =====\n'
def maxn(n, ddist):
cmf = []
for p in ddist.cmfy:
cmf.append(p**n)
tcmf = []
for p in ddist.tcmfy:
tcmf.append(p**n)
pmf = [cmf[0]]
for i in range(1, len(cmf)):
pmf.append(cmf[i] - cmf[i - 1])
if len(tcmf) == 0:
return DDist(ddist.lb, pmf, ddist.h, [], ddist.th)
else:
tpmf = [tcmf[0] - cmf[-1]]
for i in range(1, len(tcmf)):
tpmf.append(tcmf[i] - tcmf[i - 1])
return DDist(ddist.lb, pmf, ddist.h, tpmf, ddist.th)
def testCond():
print '===== test cond ====='
print '\n>>> normal\n'
p = 0.8
mu1 = 10
sigma1 = 3
mu2 = 20
sigma2 = 1
x1 = []
x2 = []
y = []
for i in range(100000):
r = np.random.random()
if r < p:
s = np.random.normal(mu1, sigma1)
x1.append(s)
y.append(s)
else:
s = np.random.normal(mu2, sigma2)
x2.append(s)
y.append(s)
ddist0 = DDist.create(x1)
ddist1 = DDist.create(x2)
ddist2 = DDist.create(y)
ddist3 = cond(p, ddist0, ddist1)
print 'base1', ddist0.mean, ddist0.std
print 'base2', ddist1.mean, ddist1.std
print 'sim', ddist2.mean, ddist2.std
print 'ana', ddist3.mean, ddist3.std
ddist2.plot('/tmp/test_cond_norm1.pdf')
ddist3.plot('/tmp/test_cond_norm2.pdf')
print '\n>>> pareto\n'
p = 0.5
a1 = 1.1
a2 = 1.2
x1 = []
x2 = []
y = []
for i in range(100000):
r = np.random.random()
if r < p:
s = np.random.pareto(a1)
x1.append(s)
y.append(s)
else:
s = np.random.pareto(a2) + 10
x2.append(s)
y.append(s)
ddist0 = DDist.create(x1, tail=('p', 0.1), tnh=10)
ddist1 = DDist.create(x2, tail=('p', 0.1), tnh=10)
ddist2 = DDist.create(y)
ddist3 = cond(p, ddist0, ddist1)
print 'base1', ddist0.mean, ddist0.std
print 'base2', ddist1.mean, ddist1.std
print 'sim', ddist2.mean, ddist2.std
print 'ana', ddist3.mean, ddist3.std
ddist2.plot('/tmp/test_cond_pareto1.pdf')
ddist3.plot('/tmp/test_cond_pareto2.pdf')
print '===== end =====\n'
def testCond():
print '===== test cond ====='
print '\n>>> normal\n'
p = 0.8
mu1 = 10
sigma1 = 3
mu2 = 20
sigma2 = 1
x1 = []
x2 = []
y = []
for i in range(100000):
r = np.random.random()
if r < p:
s = np.random.normal(mu1, sigma1)
x1.append(s)
y.append(s)
else:
s = np.random.normal(mu2, sigma2)
x2.append(s)
y.append(s)
ddist0 = DDist.create(x1)
ddist1 = DDist.create(x2)
ddist2 = DDist.create(y)
ddist3 = cond(p, ddist0, ddist1)
print 'base1', ddist0.mean, ddist0.std
print 'base2', ddist1.mean, ddist1.std
print 'sim', ddist2.mean, ddist2.std
print 'ana', ddist3.mean, ddist3.std
ddist2.plot('/tmp/test_cond_norm1.pdf')
ddist3.plot('/tmp/test_cond_norm2.pdf')
print '\n>>> pareto\n'
p = 0.5
a1 = 1.1
a2 = 1.2
x1 = []
x2 = []
y = []
for i in range(100000):
r = np.random.random()
if r < p:
s = np.random.pareto(a1)
x1.append(s)
y.append(s)
else:
s = np.random.pareto(a2) + 10
x2.append(s)
y.append(s)
ddist0 = DDist.create(x1, tail=('p', 0.1), tnh=10)
ddist1 = DDist.create(x2, tail=('p', 0.1), tnh=10)
ddist2 = DDist.create(y)
ddist3 = cond(p, ddist0, ddist1)
print 'base1', ddist0.mean, ddist0.std
print 'base2', ddist1.mean, ddist1.std
print 'sim', ddist2.mean, ddist2.std
print 'ana', ddist3.mean, ddist3.std
ddist2.plot('/tmp/test_cond_pareto1.pdf')
ddist3.plot('/tmp/test_cond_pareto2.pdf')
print '===== end =====\n'
def getEPLatencyDist(n, ddist, sync, elen):
f = int(np.ceil(n / 2.0) - 1)
rtrip = quorum(n, f, ddist)
syncRtrip = rtrip + sync
elatency = maxn(n, syncRtrip)
#wait time is a uniform distribution on elen
lb = 0
ub = elen
length = int((ub - lb) / ddist.h) + 1
pmf = [1.0 / length] * length
ewait = DDist(lb, pmf, h=ddist.h, th=ddist.th)
delay = oodelay(elatency, {
'arrival.process': 'fixed',
'fixed.interval': elen
})
return ewait + delay, delay, rtrip
def _quorumR(n, m, sTrip, rTrip):
if m <= 2:
raise ValueError(
'We only consider cases when quorum size larger than 2')
scmf, stcmf = _paddingSTrip(sTrip, rTrip.lb, rTrip.tb, rTrip.ub)
cmf = []
tcmf = []
assert len(scmf) == len(rTrip.cmfy) and len(stcmf) == len(rTrip.tcmfy), \
'scmf.l = %s == %s = r.cmfy.l, tscmf.l = %s == %s = r.tcmfy.l' %(
len(scmf), len(rTrip.cmfy), len(stcmf), len(rTrip.tcmfy))
#scmf/cmf and stcmf/tcmf are aligned
for i in range(len(scmf)):
ps = scmf[i]
pr = rTrip.cmfy[i]
p = 0.0
for j in range(m, n + 1):
for k in range(3):
if n - 2 - j + k >= 0 and j - k >= 0:
p += comb(2, k) * ps**k * (1 - ps)**(2 - k) * \
comb(n - 2, j - k) * \
pr**(j - k) * (1 - pr)**(n - 2 - j + k)
cmf.append(p)
for i in range(len(stcmf)):
ps = stcmf[i]
pr = rTrip.tcmfy[i]
p = 0.0
for j in range(m, n + 1):
for k in range(3):
if n - 2 - j + k >= 0 and j - k >= 0:
p += comb(2, k) * ps**k * (1 - ps)**(2 - k) * \
comb(n - 2, j - k) * \
pr**(j - k) * (1 - pr)**(n - 2 - j + k)
tcmf.append(p)
#compute pmf
pmf = [cmf[0]]
for i in range(1, len(cmf)):
pmf.append(cmf[i] - cmf[i - 1])
if len(tcmf) == 0:
tpmf = []
else:
tpmf = [tcmf[0] - cmf[-1]]
for i in range(1, len(tcmf)):
tpmf.append(tcmf[i] - tcmf[i - 1])
return DDist(rTrip.lb, pmf, rTrip.h, tpmf, rTrip.th)
def cond(p, ddist1, ddist2):
if ddist1.h != ddist2.h or ddist1.th != ddist2.th:
raise ValueError('DDists must have the same sample rate to use cond')
h = ddist1.h
th = ddist1.th
lb = min(ddist1.lb, ddist2.lb)
tb = max(ddist1.tb, ddist2.tb)
ub = max(ddist1.ub, ddist2.ub)
pmf = []
tpmf = []
pmf1 = _paddingPmf(ddist1, lb, tb)
pmf2 = _paddingPmf(ddist2, lb, tb)
assert len(pmf1) == len(pmf2), \
'len(pmf1) = %s == %s = len(pmf2)'%(len(pmf1), len(pmf2))
tpmf1 = _paddingTpmf(ddist1, tb, ub)
tpmf2 = _paddingTpmf(ddist2, tb, ub)
assert len(tpmf1) == len(tpmf2), \
'len(tpmf1) = %s == %s = len(tpmf2)'%(len(tpmf1), len(tpmf2))
for i in range(len(pmf1)):
pmf.append(p * pmf1[i] + (1 - p) * pmf2[i])
for i in range(len(tpmf1)):
tpmf.append(p * tpmf1[i] + (1 - p) * tpmf2[i])
return DDist(lb, pmf, h, tpmf, th)
def _quorumC(n, m, rTrip):
#front distribution
cmf = [0.0] * len(rTrip.cmfy)
for i in range(len(cmf)):
p = rTrip.cmfy[i]
for j in range(m, n + 1):
cmf[i] += comb(n, j) * p**j * (1 - p)**(n - j)
#tail disctribution
tcmf = [0.0] * len(rTrip.tcmfy)
for i in range(len(tcmf)):
p = rTrip.tcmfy[i]
for j in range(m, n + 1):
tcmf[i] += comb(n, j) * p**j * (1 - p)**(n - j)
#comput pmf
pmf = [cmf[0]]
for i in range(1, len(cmf)):
pmf.append(cmf[i] - cmf[i - 1])
if len(tcmf) == 0:
tpmf = []
else:
tpmf = [tcmf[0] - cmf[-1]]
for i in range(1, len(tcmf)):
tpmf.append(tcmf[i] - tcmf[i - 1])
return DDist(rTrip.lb, pmf, rTrip.h, tpmf, rTrip.th)
def PBgA(a):
if a == "a1":
return DDist({"b1": 0.7, "b2": 0.3})
else:
return DDist({"b1": 0.2, "b2": 0.8})
def testQuorum():
print '===== test quorum ====='
lambd = 50
a = 1.8
n = 5
f = 2
x = {'gen' : [], 'exp':[], 'pareto':[]}
yc = {'gen' : [], 'exp':[], 'pareto':[]}
yr = {'gen' : [], 'exp':[], 'pareto':[]}
y = {'gen' : [], 'exp':[], 'pareto':[]}
dists = ['gen', 'exp', 'pareto']
leader = 0
def gen():
r = np.random.random()
if r < 0.2:
return 1
elif r < 0.4:
return 2
elif r < 0.6:
return 3
elif r < 0.8:
return 4
else:
return 8
for i in range(100000):
first = {}
second = {}
latencies = {'gen':[], 'exp':[], 'pareto':[]}
learner = np.random.randint(n)
for acceptor in range(n):
if leader == acceptor:
for k in dists:
first[k] = 0
else:
first['gen'] = gen()
first['exp'] = np.random.exponential(lambd)
first['pareto'] = np.random.pareto(a)
for k in dists:
x[k].append(first[k])
if acceptor == learner:
for k in dists:
second[k] = 0
else:
second['gen'] = gen()
second['exp'] = np.random.exponential(lambd)
second['pareto'] = np.random.pareto(a)
for k in dists:
x[k].append(second[k])
for k in dists:
latencies[k].append(first[k] + second[k])
for k in dists:
l = sorted(latencies[k])
y[k].append(l[n - f - 1])
if leader == learner:
assert l[0] == 0
yc[k].append(l[n - f - 1])
else:
yr[k].append(l[n - f - 1])
for k in dists:
print '\n>>> %s\n'%k
if k == 'gen':
ddist0 = DDist.create(x[k], h=1, tail=('b', 4), tnh=4)
elif k == 'exp':
ddist0 = DDist.create(x[k], h=0.5)
else:
ddist0 = DDist.create(x[k], h=0.5, tail=('p', 0.1), tnh=20)
print 'created ddist0'
if k == 'gen':
ddist1 = DDist.create(y[k], h=1, tail=('b', 8), tnh=4)
ddistc1 = DDist.create(yc[k], h=1, tail=('b', 8), tnh=4)
ddistr1 = DDist.create(yr[k], h=1, tail=('b', 8), tnh=4)
print 'base pmf', ddist1.getPmfxy()
print 'base c pmf', ddistc1.getPmfxy()
print 'base r pmf', ddistr1.getPmfxy()
else:
ddist1 = DDist.create(y[k], h=0.05)
ddistc1 = DDist.create(yc[k], h=0.05)
ddistr1 = DDist.create(yr[k], h=0.05)
ddist2 = quorum(n, f, ddist0)
print 'computed quorum'
rtrip = ddist0 + ddist0
ddistc2 = _quorumC(n - 1, n - f - 1, rtrip)
ddistr2 = _quorumR(n, n - f, ddist0, rtrip)
if k == 'gen':
print 'q pmf', ddist2.getPmfxy()
print 'q c pmf', ddistc2.getPmfxy()
print 'q r pmf', ddistr2.getPmfxy()
print 'latency', ddist0.mean, ddist0.std
print 'quorum sim', ddist1.mean, ddist1.std
print 'quorum c sim', ddistc1.mean, ddistc1.std
print 'quorum r sim', ddistr1.mean, ddistr1.std
print 'quorum ana', ddist2.mean, ddist2.std
print 'quorum c ana', ddistc2.mean, ddistc2.std
print 'quorum r ana', ddistr2.mean, ddistr2.std
ddist1.plot('/tmp/test_quorum_%s1.pdf'%k)
ddistc1.plot('/tmp/test_quorumc_%s1.pdf'%k)
ddistr1.plot('/tmp/test_quorumr_%s1.pdf'%k)
ddist2.plot('/tmp/test_quorum_%s2.pdf'%k)
ddistc2.plot('/tmp/test_quorumc_%s2.pdf'%k)
ddistr2.plot('/tmp/test_quorumr_%s2.pdf'%k)
print '===== end =====\n'
def flips(nFlips):
'''nFlips число бросаний монеты,
возвращает h - частоту выпадения орла.'''
res = choices([0, 1], k=nFlips)
h = sum(res) / nFlips #sum(res) число выпадений орла
return h
res = [flips(10) for k in range(8)]
print(res)
res = [flips(1000) for k in range(8)]
print(res)
from ddist import DDist, JDist
#H орёл, T - решка.
dm = DDist({"H": 1 / 2, "T": 1 / 2})
print(dm.trials(8))
PAB = DDist({
("H", "H"): 1 / 4,
("H", "T"): 1 / 4,
("T", "H"): 1 / 4,
("T", "T"): 1 / 4
})
PA = PAB.marginalizeOut(1)
print(PA)
def pTestGivenDis(d):
if d == "болен":
return DDist({"положительный": 0.99, "отрицательный": 0.01})
elif d == "здоров":
def testoodelay():
print '===== test oodelay =====\n'
intvl = 30
def genexpointvl():
return np.random.exponential(intvl)
def genfixedintvl():
return intvl
lambd = 100
a = 1.8
m = 100
lb = 50
tb = 150
ub = 1000
def customlatency():
r = np.random.random()
if r < 0.2:
return 30
elif r < 0.4:
return 60
elif r < 0.6:
return 90
elif r < 0.8:
return 120
else:
return 180
def expolatency():
return np.random.exponential(lambd)
def paretolatency():
return np.random.pareto(a) + m
def uniformlatency():
r = np.random.random()
if r < 0.8:
rr = np.random.random()
return lb + rr * (tb - lb)
else:
rr = np.random.random()
return tb + rr * (ub - tb)
intvlkeys = {'fixed' : genfixedintvl,
'poisson' : genexpointvl}
intvlopts = {'fixed' : 'fixed.interval',
'poisson' : 'poisson.lambda'}
latencykeys = {'custom' : customlatency,
'expo' : expolatency,
'pareto' : paretolatency,
'uniform' : uniformlatency,
}
for ik in intvlkeys:
for lk in latencykeys:
print '\n>>> %s, %s\n'%(ik, lk)
x, y = runoodelay(intvlkeys[ik], latencykeys[lk])
if lk == 'custom':
ddist0 = DDist.create(x, h=30, tail=('b', 120), tnh=2)
elif lk == 'expo':
ddist0 = DDist.create(x, h=0.5)
elif lk == 'uniform':
ddist0 = DDist.create(x, h=0.5, tail=('b', 150), tnh=10)
else:
ddist0 = DDist.create(x, h=0.5, tail=('p', 0.1), tnh=10)
if lk == 'custom':
ddist1 = DDist.create(y, h=30)
else:
ddist1 = DDist.create(y, h=0.5)
if ik == 'fixed':
ddist2 = oodelay(ddist0, {'arrival.process' : 'fixed',
'fixed.interval' : intvl})
else:
ddist2 = oodelay(ddist0, {'arrival.process' : 'poisson',
'poisson.lambda' : 1.0 / intvl})
print 'sim', ddist1.mean, ddist1.std
print 'ana', ddist2.mean, ddist2.std
if lk == 'custom':
print 'sim pmf', ddist1.getPmfxy()
print 'ana pmf', ddist2.getPmfxy()
print '===== end =====\n'
def testoodelay():
print '===== test oodelay =====\n'
intvl = 30
def genexpointvl():
return np.random.exponential(intvl)
def genfixedintvl():
return intvl
lambd = 100
a = 1.8
m = 100
lb = 50
tb = 150
ub = 1000
def customlatency():
r = np.random.random()
if r < 0.2:
return 30
elif r < 0.4:
return 60
elif r < 0.6:
return 90
elif r < 0.8:
return 120
else:
return 180
def expolatency():
return np.random.exponential(lambd)
def paretolatency():
return np.random.pareto(a) + m
def uniformlatency():
r = np.random.random()
if r < 0.8:
rr = np.random.random()
return lb + rr * (tb - lb)
else:
rr = np.random.random()
return tb + rr * (ub - tb)
intvlkeys = {'fixed': genfixedintvl, 'poisson': genexpointvl}
intvlopts = {'fixed': 'fixed.interval', 'poisson': 'poisson.lambda'}
latencykeys = {
'custom': customlatency,
'expo': expolatency,
'pareto': paretolatency,
'uniform': uniformlatency,
}
for ik in intvlkeys:
for lk in latencykeys:
print '\n>>> %s, %s\n' % (ik, lk)
x, y = runoodelay(intvlkeys[ik], latencykeys[lk])
if lk == 'custom':
ddist0 = DDist.create(x, h=30, tail=('b', 120), tnh=2)
elif lk == 'expo':
ddist0 = DDist.create(x, h=0.5)
elif lk == 'uniform':
ddist0 = DDist.create(x, h=0.5, tail=('b', 150), tnh=10)
else:
ddist0 = DDist.create(x, h=0.5, tail=('p', 0.1), tnh=10)
if lk == 'custom':
ddist1 = DDist.create(y, h=30)
else:
ddist1 = DDist.create(y, h=0.5)
if ik == 'fixed':
ddist2 = oodelay(ddist0, {
'arrival.process': 'fixed',
'fixed.interval': intvl
})
else:
ddist2 = oodelay(ddist0, {
'arrival.process': 'poisson',
'poisson.lambda': 1.0 / intvl
})
print 'sim', ddist1.mean, ddist1.std
print 'ana', ddist2.mean, ddist2.std
if lk == 'custom':
print 'sim pmf', ddist1.getPmfxy()
print 'ana pmf', ddist2.getPmfxy()
print '===== end =====\n'
def testQuorum():
print '===== test quorum ====='
lambd = 50
a = 1.8
n = 5
f = 2
x = {'gen': [], 'exp': [], 'pareto': []}
yc = {'gen': [], 'exp': [], 'pareto': []}
yr = {'gen': [], 'exp': [], 'pareto': []}
y = {'gen': [], 'exp': [], 'pareto': []}
dists = ['gen', 'exp', 'pareto']
leader = 0
def gen():
r = np.random.random()
if r < 0.2:
return 1
elif r < 0.4:
return 2
elif r < 0.6:
return 3
elif r < 0.8:
return 4
else:
return 8
for i in range(100000):
first = {}
second = {}
latencies = {'gen': [], 'exp': [], 'pareto': []}
learner = np.random.randint(n)
for acceptor in range(n):
if leader == acceptor:
for k in dists:
first[k] = 0
else:
first['gen'] = gen()
first['exp'] = np.random.exponential(lambd)
first['pareto'] = np.random.pareto(a)
for k in dists:
x[k].append(first[k])
if acceptor == learner:
for k in dists:
second[k] = 0
else:
second['gen'] = gen()
second['exp'] = np.random.exponential(lambd)
second['pareto'] = np.random.pareto(a)
for k in dists:
x[k].append(second[k])
for k in dists:
latencies[k].append(first[k] + second[k])
for k in dists:
l = sorted(latencies[k])
y[k].append(l[n - f - 1])
if leader == learner:
assert l[0] == 0
yc[k].append(l[n - f - 1])
else:
yr[k].append(l[n - f - 1])
for k in dists:
print '\n>>> %s\n' % k
if k == 'gen':
ddist0 = DDist.create(x[k], h=1, tail=('b', 4), tnh=4)
elif k == 'exp':
ddist0 = DDist.create(x[k], h=0.5)
else:
ddist0 = DDist.create(x[k], h=0.5, tail=('p', 0.1), tnh=20)
print 'created ddist0'
if k == 'gen':
ddist1 = DDist.create(y[k], h=1, tail=('b', 8), tnh=4)
ddistc1 = DDist.create(yc[k], h=1, tail=('b', 8), tnh=4)
ddistr1 = DDist.create(yr[k], h=1, tail=('b', 8), tnh=4)
print 'base pmf', ddist1.getPmfxy()
print 'base c pmf', ddistc1.getPmfxy()
print 'base r pmf', ddistr1.getPmfxy()
else:
ddist1 = DDist.create(y[k], h=0.05)
ddistc1 = DDist.create(yc[k], h=0.05)
ddistr1 = DDist.create(yr[k], h=0.05)
ddist2 = quorum(n, f, ddist0)
print 'computed quorum'
rtrip = ddist0 + ddist0
ddistc2 = _quorumC(n - 1, n - f - 1, rtrip)
ddistr2 = _quorumR(n, n - f, ddist0, rtrip)
if k == 'gen':
print 'q pmf', ddist2.getPmfxy()
print 'q c pmf', ddistc2.getPmfxy()
print 'q r pmf', ddistr2.getPmfxy()
print 'latency', ddist0.mean, ddist0.std
print 'quorum sim', ddist1.mean, ddist1.std
print 'quorum c sim', ddistc1.mean, ddistc1.std
print 'quorum r sim', ddistr1.mean, ddistr1.std
print 'quorum ana', ddist2.mean, ddist2.std
print 'quorum c ana', ddistc2.mean, ddistc2.std
print 'quorum r ana', ddistr2.mean, ddistr2.std
ddist1.plot('/tmp/test_quorum_%s1.pdf' % k)
ddistc1.plot('/tmp/test_quorumc_%s1.pdf' % k)
ddistr1.plot('/tmp/test_quorumr_%s1.pdf' % k)
ddist2.plot('/tmp/test_quorum_%s2.pdf' % k)
ddistc2.plot('/tmp/test_quorumc_%s2.pdf' % k)
ddistr2.plot('/tmp/test_quorumr_%s2.pdf' % k)
print '===== end =====\n'