def optimize(self, calc_ste=False, Ifilter=None, bounds=None, verbose=True,
opts={}, *args, **kw_args):
# logger.info('Marginal likelihood optimization.')
if verbose:
print 'Marginal likelihood optimization.'
t0 = time.time()
conv, info = OPT.opt_hyper(self, Ifilter=Ifilter, bounds=bounds,
opts=opts, *args, **kw_args)
t1 = time.time()
# if logger.levelno == logger.DEBUG:
if verbose:
# logger.debug('Time elapsed: %.2fs', t1-t0)
print 'Converged:', conv
print 'Time elapsed: %.2f s' % (t1-t0)
grad = self.LML_grad()
grad_norm = 0
for key in grad.keys():
grad_norm += (grad[key]**2).sum()
grad_norm = sp.sqrt(grad_norm)
print 'Log Marginal Likelihood: %.7f.' % self.LML()
print 'Gradient norm: %.7f.' % grad_norm
# logger.debug('Log Marginal Likelihood: %.7f.', self.LML())
# logger.debug('Gradient norm: %.7f.', grad_norm)
if calc_ste:
self.calc_ste(verbose=verbose)
return conv, info
def optimize(self, calc_ste=False, Ifilter=None, bounds=None, verbose=True,
opts={}, *args, **kw_args):
# logger.info('Marginal likelihood optimization.')
if verbose:
print('Marginal likelihood optimization.')
t0 = time.time()
conv, info = OPT.opt_hyper(self, Ifilter=Ifilter, bounds=bounds,
opts=opts, *args, **kw_args)
t1 = time.time()
# if logger.levelno == logger.DEBUG:
if verbose:
# logger.debug('Time elapsed: %.2fs', t1-t0)
print(('Converged:', conv))
print(('Time elapsed: %.2f s' % (t1-t0)))
grad = self.LML_grad()
grad_norm = 0
for key in list(grad.keys()):
grad_norm += (grad[key]**2).sum()
grad_norm = sp.sqrt(grad_norm)
print(('Log Marginal Likelihood: %.7f.' % self.LML()))
print(('Gradient norm: %.7f.' % grad_norm))
# logger.debug('Log Marginal Likelihood: %.7f.', self.LML())
# logger.debug('Gradient norm: %.7f.', grad_norm)
if calc_ste:
self.calc_ste(verbose=verbose)
return conv, info
# compare with mtSet implementation
params = {}
params['Cg'] = SP.randn(int(0.5*P*(P+1)))
params['Cn'] = SP.randn(int(0.5*P*(P+1)))
print "check gradient with gp2kronSum"
gp = gp2kronSum(mu,Cg,Cn,XX)
gp.setParams(params)
if 0:
gp.set_reml(False)
print "test optimization"
start = TIME.time()
conv,info = OPT.opt_hyper(gp,params,factr=1e3)
print 'Reml GP:', TIME.time()-start
if mtSet_present:
params1 = copy.copy(params)
params1['mean'] = SP.zeros(mu1.getParams().shape[0])
gp1 = gp2kronSumMtSet(mu1,Cg1,Cn1,XX)
gp1.setParams(params1)
start = TIME.time()
conv1,info = OPT.opt_hyper(gp1,params1,factr=1e3)
print 'Old GP:', TIME.time()-start
print conv
ipdb.set_trace()
gp.checkGradient()
if 0:
gp.K.optimizeAB(n=100)
print(('a:', gp.K.a))
print(('b:', gp.K.b))
ipdb.set_trace()
gp.K.optimizeABgrad()
for i in range(10):
ipdb.set_trace()
conv,info = OPT.opt_hyper(gp,params,factr=1e-3)
print(conv)
print('C1')
print((C1.K()))
print('C2')
print((C2.K()))
print('Cn')
print((Cn.K()))
ipdb.set_trace()
print(('before a and b opt:', gp.LML()))
gp.K.optimizeAB(n=100)
print(('after a and b opt:', gp.LML()))
print(('a:', gp.K.a))
print(('b:', gp.K.b))
print "creating gp2kronSum object"
XX = SP.dot(X, X.T)
XX /= XX.diagonal().mean()
gp = gp2kronSum(mu, Cg, Cn, XX)
gp.setParams(params)
if "ML" in sys.argv:
print "ML estimation"
gp.set_reml(False)
else:
print "REML estimation"
print "optimization of GP parameters"
start = TIME.time()
conv, info = OPT.opt_hyper(gp, params, factr=1e3)
print 'time for fitting GP:', TIME.time() - start
print conv
print "creating lmm for association using GP object"
assoc = lmm.LmmKronecker(gp=gp)
#test snps
print "testing SNPs with any effect"
pv, LL_snps, LL_snps_0 = assoc.test_snps(snps)
if 1:
print "forward selection step"
print "adding SNP with smalles pv as fixed effect"
i_pv = pv.argsort()
# compare with mtSet implementation
params = {}
params['Cg'] = SP.randn(int(0.5*P*(P+1)))
params['Cn'] = SP.randn(int(0.5*P*(P+1)))
print("check gradient with gp2kronSum")
gp = gp2kronSum(mu,Cg,Cn,XX)
gp.setParams(params)
if 0:
gp.set_reml(False)
print("test optimization")
start = TIME.time()
conv,info = OPT.opt_hyper(gp,params,factr=1e3)
print(('Reml GP:', TIME.time()-start))
if mtSet_present:
params1 = copy.copy(params)
params1['mean'] = SP.zeros(mu1.getParams().shape[0])
gp1 = gp2kronSumMtSet(mu1,Cg1,Cn1,XX)
gp1.setParams(params1)
start = TIME.time()
conv1,info = OPT.opt_hyper(gp1,params1,factr=1e3)
print(('Old GP:', TIME.time()-start))
print(conv)
ipdb.set_trace()
print("creating gp2kronSum object")
XX = SP.dot(X,X.T)
XX/= XX.diagonal().mean()
gp = gp2kronSum(mu,Cg,Cn,XX)
gp.setParams(params)
if "ML" in sys.argv:
print("ML estimation")
gp.set_reml(False)
else:
print("REML estimation")
print("optimization of GP parameters")
start = TIME.time()
conv,info = OPT.opt_hyper(gp,params,factr=1e3)
print(('time for fitting GP:', TIME.time()-start))
print(conv)
print("creating lmm for association using GP object")
assoc = lmm.LmmKronecker(gp=gp)
#test snps
print("testing SNPs with any effect")
pv,LL_snps,LL_snps_0= assoc.test_snps(snps)
if 1:
gp.checkGradient()
if 0:
gp.K.optimizeAB(n=100)
print 'a:', gp.K.a
print 'b:', gp.K.b
ipdb.set_trace()
gp.K.optimizeABgrad()
for i in range(10):
ipdb.set_trace()
conv,info = OPT.opt_hyper(gp,params,factr=1e-3)
print conv
print 'C1'
print C1.K()
print 'C2'
print C2.K()
print 'Cn'
print Cn.K()
ipdb.set_trace()
print 'before a and b opt:', gp.LML()
gp.K.optimizeAB(n=100)
print 'after a and b opt:', gp.LML()
print 'a:', gp.K.a
print 'b:', gp.K.b