def optimize(self,params0):
""" initialize """
self.gp.setParams(params0)
self.cache['params0'] = params0
self.cache['lml0'] = self.gp.LML()
self.cache['lmlGrad0'] = self.gp.LMLgrad()
""" optimize """
gpopt = limix.CGPopt(self.gp)
conv = gpopt.opt()
""" store stuff """
self.cache['lml'] = self.gp.LML()
self.cache['lmlGrad'] = self.gp.LMLgrad()
self.cache['params'] = self.gp.getParams()
self.cache['X'] = self.Ca.getParams().reshape((self.N,self.k),order='F')
self.cache['K'] = self.Ca.K()
if self.interaction:
self.cache['Ki'] = self.Ca.K()*self.Cb2.K()
else:
self.cache['Ki'] = None
self.cache['var'] = {}
self.cache['var']['K0'] = self.C0.getParams()[0]**2
self.cache['var']['K'] = self.cache['K'].diagonal().mean()
if self.interaction:
self.cache['var']['Ki'] = self.cache['Ki'].diagonal().mean()
self.cache['var']['noise'] = self.ll.getParams()[0]**2
self.cache['K'] /= self.cache['var']['K']
if self.interaction:
self.cache['Ki'] /= self.cache['var']['Ki']
return conv
def fit(self, Params0=None, grad_threshold=1e-2):
"""
fit a variance component model with the predefined design and the initialization and returns all the results
"""
# GPVD initialization
lik = limix.CLikNormalNULL()
# Initial Params
if Params0 == None:
n_params = self.C1.getNumberParams()
n_params += self.C2.getNumberParams()
Params0 = SP.rand(n_params)
# MultiGP Framework
covar = []
gp = []
mean = []
for i in range(self.N):
covar.append(limix.CLinCombCF())
covar[i].addCovariance(self.C1)
covar[i].addCovariance(self.C2)
coeff = SP.array([self.eigen[i], 1])
covar[i].setCoeff(coeff)
mean.append(limix.CLinearMean(self.Yt[:, i], SP.eye(self.P)))
gpVD = limix.CGPvarDecomp(covar[0], lik, mean[0], SP.ones(self.N),
self.P, self.Yt, Params0)
for i in range(self.N):
gp.append(limix.CGPbase(covar[i], lik, mean[i]))
gp[i].setY(self.Yt[:, i])
gpVD.addGP(gp[i])
# Optimization
gpVD.initGPs()
gpopt = limix.CGPopt(gpVD)
LML0 = -1.0 * gpVD.LML()
start_time = time.time()
conv = gpopt.opt()
time_train = time.time() - start_time
LML = -1.0 * gpVD.LML()
LMLgrad = SP.linalg.norm(gpVD.LMLgrad()['covar'])
Params = gpVD.getParams()['covar']
# Check whether limix::CVarianceDecomposition.train() has converged
if conv != True or LMLgrad > grad_threshold or Params.max() > 10:
print 'limix::CVarianceDecomposition::train has not converged'
res = None
else:
res = {
'Params0': Params0,
'Params': Params,
'LML': SP.array([LML]),
'LML0': SP.array([LML0]),
'LMLgrad': SP.array([LMLgrad]),
'time_train': SP.array([time_train]),
}
return res
pass
def test_fit(self):
#create optimization object
self.gpopt = limix.CGPopt(self.gp)
#run
RV = self.gpopt.opt()
RV = self.gpopt.opt()
RV = RV & (SP.absolute(self.gp.LMLgrad()['X']).max() < 1E-1)
RV = RV & (SP.absolute(self.gp.LMLgrad()['covar']).max() < 1E-1)
RV = RV & (SP.absolute(self.gp.LMLgrad()['lik']).max() < 1E-1)
self.assertTrue(RV)
def test_fit(self):
#create optimization object
self.gpopt = limix.CGPopt(self.gp)
self.gpopt.setOptBoundLower(self.constrainL)
self.gpopt.setOptBoundUpper(self.constrainU)
#run
self.gpopt.opt()
params = SP.concatenate(
(self.gp.getParams()['covar'], self.gp.getParams()['lik']))[:, 0]
params_true = SP.array(
[0.28822188, 0.35271548, 0.13709146, 0.49447424])
RV = ((params - params_true)**2).max() < 1e-6
RV = RV & (SP.absolute(self.gp.LMLgrad()['lik']).max() < 1E-4)
RV = RV & (SP.absolute(self.gp.LMLgrad()['covar']).max() < 1E-4)
self.assertTrue(RV)
GP['covar'].addCovariance(GP['covar_E']) #liklihood: gaussian GP['ll'] = limix.CLikNormalIso() GP['data'] = limix.CData() GP['hyperparams'] = limix.CGPHyperParams() #Create GP instance GP['gp'] = limix.CGPbase(GP['data'], GP['covar'], GP['ll']) #set data GP['gp'].setY(Y) #input: effectively we require the group for each sample (CCovFreeform requires this) Xtrain = SP.zeros([Y.shape[0], 1]) Xtrain[N::1] = 1 GP['gp'].setX(Xtrain) gpopt = limix.CGPopt(GP['gp']) #constraints: make sure that noise level does not go completel crazy constrainU = limix.CGPHyperParams() constrainL = limix.CGPHyperParams() constrainU['lik'] = +5 * SP.ones([1]) constrainL['lik'] = -5 * SP.ones([1]) gpopt.setOptBoundLower(constrainL) gpopt.setOptBoundUpper(constrainU) #filters: we do not need to optimize the overall scaling prameter in front of Kpop P = [] L = [] #use a number of restarts... #TODO: think about a more elegant scheme to handle multiple restarts properly t0 = time.time() for i in xrange(1):
def train(self,rank=20,Kpop=True,LinearARD=False):
"""train panama module"""
if 0:
covar = limix.CCovLinearISO(rank)
ll = limix.CLikNormalIso()
X0 = sp.random.randn(self.N,rank)
X0 = PCA(self.Y,rank)[0]
X0 /= sp.sqrt(rank)
covar_params = sp.array([1.0])
lik_params = sp.array([1.0])
hyperparams = limix.CGPHyperParams()
hyperparams['covar'] = covar_params
hyperparams['lik'] = lik_params
hyperparams['X'] = X0
constrainU = limix.CGPHyperParams()
constrainL = limix.CGPHyperParams()
constrainU['covar'] = +5*sp.ones_like(covar_params);
constrainL['covar'] = 0*sp.ones_like(covar_params);
constrainU['lik'] = +5*sp.ones_like(lik_params);
constrainL['lik'] = 0*sp.ones_like(lik_params);
if 1:
covar = limix.CSumCF()
if LinearARD:
covar_1 = limix.CCovLinearARD(rank)
covar_params = []
for d in range(rank):
covar_params.append(1/sp.sqrt(d+2))
else:
covar_1 = limix.CCovLinearISO(rank)
covar_params = [1.0]
covar.addCovariance(covar_1)
if self.use_Kpop:
covar_2 = limix.CFixedCF(self.Kpop)
covar.addCovariance(covar_2)
covar_params.append(1.0)
ll = limix.CLikNormalIso()
X0 = PCA(self.Y,rank)[0]
X0 /= sp.sqrt(rank)
covar_params = sp.array(covar_params)
lik_params = sp.array([1.0])
hyperparams = limix.CGPHyperParams()
hyperparams['covar'] = covar_params
hyperparams['lik'] = lik_params
hyperparams['X'] = X0
constrainU = limix.CGPHyperParams()
constrainL = limix.CGPHyperParams()
constrainU['covar'] = +5*sp.ones_like(covar_params);
constrainL['covar'] = -5*sp.ones_like(covar_params);
constrainU['lik'] = +5*sp.ones_like(lik_params);
gp=limix.CGPbase(covar,ll)
gp.setY(self.Y)
gp.setX(X0)
lml0 = gp.LML(hyperparams)
dlml0 = gp.LMLgrad(hyperparams)
gpopt = limix.CGPopt(gp)
gpopt.setOptBoundLower(constrainL);
gpopt.setOptBoundUpper(constrainU);
t1 = time.time()
gpopt.opt()
t2 = time.time()
#Kpanama
self.Xpanama = covar_1.getX()
if LinearARD:
self.Xpanama /= self.Xpanama.std(0)
self.Kpanama = covar_1.K()
self.Kpanama/= self.Kpanama.diagonal().mean()
# Ktot
self.Ktot = covar_1.K()
if self.use_Kpop:
self.Ktot += covar_2.K()
self.Ktot/= self.Ktot.diagonal().mean()
#store variances
V = {}
if LinearARD:
V['LinearARD'] = covar_1.getParams()**2*covar_1.getX().var(0)
else:
V['Kpanama'] = sp.array([covar_1.K().diagonal().mean()])
if self.use_Kpop:
V['Kpop'] = sp.array([covar_2.K().diagonal().mean()])
V['noise'] = gp.getParams()['lik']**2
self.varianceComps = V
covar_r = limix.CFixedCF(SP.eye(N))
Xr = SP.zeros([N, 0])
Xc = SP.zeros([D, 0])
gp = limix.CGPkronecker(covar_r, covar_c)
gp.setX_r(Xr)
gp.setX_c(Xc)
gp.setY(Y)
params = limix.CGPHyperParams()
params["covar_r"] = SP.zeros([covar_r.getNumberParams()])
params["covar_c"] = SP.zeros([covar_c.getNumberParams()])
params["lik"] = SP.log([0.1, 0.1])
gp.setParams(params)
opt = limix.CGPopt(gp)
opt.opt()
#opt.gradCheck()
if 0:
opt_params = limix.CGPHyperParams()
opt_params["lik"] = params["lik"]
gp.setParams(opt_params)
pdb.set_trace()
#lmm object
lmm = limix.CGPLMM(gp)
lmm.setSNPs(S)
lmm.setCovs(SP.ones([N, 1]))
lmm.setPheno(Y)
# initalization
params = limix.CGPHyperParams()
params['covarc1'] = SP.randn(covarc1.getNumberParams())
params['covarc2'] = SP.randn(covarc2.getNumberParams())
params['covarr1'] = SP.ones(1)
params['covarr2'] = SP.ones(1)
params['dataTerm'] = SP.randn(P + 1, 1)
gpKS.setParams(params)
# Paramask
paramMask = limix.CGPHyperParams()
paramMask['covarc1'] = covarc1.getParamMask()
paramMask['covarc2'] = covarc2.getParamMask()
paramMask['covarr1'] = covarr1.getParamMask()
paramMask['covarr2'] = covarr2.getParamMask()
# optimisation
gpKSopt = limix.CGPopt(gpKS)
start_time = time.time()
gpKSopt.opt()
elapsed_timeFast = time.time() - start_time
print "time elapsed(s): " + str(elapsed_timeFast)
paramsOpt = gpKS.getParams()
print ""
print "optimised parameters"
print SP.concatenate((paramsOpt['covarc1'], paramsOpt['covarr1'],
paramsOpt['covarc2'], paramsOpt['covarr2']))
print paramsOpt['dataTerm']
print ""
print "GP BASE"
# NormalGP
