def addmins(G,X,Y,S,D,xmin,mode=OFFHESSZERO, GRADNOISE=1e-9,EP_SOFTNESS=1e-9,EPROP_LOOPS=20):
dim=X.shape[1]
#grad elements are zero
Xg = sp.vstack([xmin]*dim)
Yg = sp.zeros([dim,1])
Sg = sp.ones([dim,1])*GRADNOISE
Dg = [[i] for i in range(dim)]
#offdiag hessian elements
nh = ((dim-1)*dim)/2
Xh = sp.vstack([sp.empty([0,dim])]+[xmin]*nh)
Dh=[]
for i in xrange(dim):
for j in xrange(i):
Dh.append([i,j])
class MJMError(Exception):
pass
if mode==OFFHESSZERO:
Yh = sp.zeros([nh,1])
Sh = sp.ones([nh,1])*GRADNOISE
elif mode==OFFHESSINFER:
raise MJMError("addmins with mode offhessinfer not implemented yet")
else:
raise MJMError("invalid mode in addmins")
#diag hessian and min
Xd = sp.vstack([xmin]*(dim+1))
Dd = [[sp.NaN]]+[[i,i] for i in xrange(dim)]
[m,V] = G.infer_full_post(Xd,Dd)
for i in xrange(dim):
if V[i,i]<0:
class MJMError(Exception):
pass
print [m,V]
raise MJMError('negative on diagonal')
yminarg = sp.argmin(Y)
Y_ = sp.array([Y[yminarg,0]]+[0.]*dim)
Z = sp.array([-1]+[1.]*dim)
F = sp.array([S[yminarg,0]]+[EP_SOFTNESS]*dim)
[Yd,Stmp] = eprop.expectation_prop(m,V,Y_,Z,F,EPROP_LOOPS)
Sd = sp.diagonal(Stmp).flatten()
Sd.resize([dim+1,1])
Yd.resize([dim+1,1])
#concat the obs
Xo = sp.vstack([X,Xg,Xd,Xh])
Yo = sp.vstack([Y,Yg,Yd,Yh])
So = sp.vstack([S,Sg,Sd,Sh])
Do = D+Dg+Dd+Dh
return [Xo,Yo,So,Do]
import eprop
import scipy as sp
from scipy import linalg as spl
from scipy import stats as sps
from matplotlib import pyplot as plt
#basic ep test on three values
m = sp.array([1.,2.,0.5])
v = sp.array([[1.,0.5,0.2],[0.5,2.,0.5],[0.2,0.5,1.]])
n=len(m)
Y = sp.array([1.,0.,1.])
Z = sp.array([-1,1,-1])
F = sp.array([0.25**2,0.,0.])
mt,vt = eprop.expectation_prop(m,v,Y,Z,F,35)
mu,vu = eprop.gaussian_fusion(m,mt,v,vt)
f,ax = plt.subplots(n,sharex=True)
ns=200
sup = sp.linspace(-6,6,ns)
for i in xrange(n):
ax[i].plot(sup,sps.norm.pdf(sup,loc=m[i],scale=sp.sqrt(v[i,i])),'b')
ax[i].plot(sup,sps.norm.pdf(sup,loc=mu[i],scale=sp.sqrt(vu[i,i])),'r')
ax[i].twinx().plot(sup,sps.norm.cdf(Z[i]*(sup-Y[i])/max(F[i],1e-20))*sps.norm.pdf(sup,loc=m[i],scale=sp.sqrt(v[i,i])),'g')
#est on a gp
import ESutils
import GPdc
nt=5
#!/usr/bin/env python2
# encoding: UTF-8
# To change this license header, choose License Headers in Project Properties.
# To change this template file, choose Tools | Templates
# and open the template in the editor.
import OPTutils
import scipy as sp
from matplotlib import pyplot as plt
import GPdc
import eprop
import pickle
[m0, V0, Y, Z, F, z] = pickle.load(open("epkill.p", "rb"))
print m0
print V0
print Y
print Z
print F
print z
import eprop
eprop.expectation_prop(m0, V0, Y, Z, F, z)
import eprop
import scipy as sp
from scipy import linalg as spl
from scipy import stats as sps
from matplotlib import pyplot as plt
# basic ep test on three values
m = sp.array([1.0, 2.0, 0.5])
v = sp.array([[1.0, 0.5, 0.2], [0.5, 2.0, 0.5], [0.2, 0.5, 1.0]])
n = len(m)
Y = sp.array([1.0, 0.0, 1.0])
Z = sp.array([-1, 1, -1])
F = sp.array([0.25 ** 2, 0.0, 0.0])
mt, vt = eprop.expectation_prop(m, v, Y, Z, F, 35)
mu, vu = eprop.gaussian_fusion(m, mt, v, vt)
f, ax = plt.subplots(n, sharex=True)
ns = 200
sup = sp.linspace(-6, 6, ns)
for i in xrange(n):
ax[i].plot(sup, sps.norm.pdf(sup, loc=m[i], scale=sp.sqrt(v[i, i])), "b")
ax[i].plot(sup, sps.norm.pdf(sup, loc=mu[i], scale=sp.sqrt(vu[i, i])), "r")
ax[i].twinx().plot(
sup,
sps.norm.cdf(Z[i] * (sup - Y[i]) / max(F[i], 1e-20)) * sps.norm.pdf(sup, loc=m[i], scale=sp.sqrt(v[i, i])),
"g",
)
# est on a gp
#!/usr/bin/env python2
#encoding: UTF-8
# To change this license header, choose License Headers in Project Properties.
# To change this template file, choose Tools | Templates
# and open the template in the editor.
import OPTutils
import scipy as sp
from matplotlib import pyplot as plt
import GPdc
import eprop
import pickle
[m0,V0,Y,Z,F,z] = pickle.load(open('epkill.p','rb'))
print m0
print V0
print Y
print Z
print F
print z
import eprop
eprop.expectation_prop(m0,V0,Y,Z,F,z)