#print (random)
#print (gvec)
#gvec=np.zeros(len(sim_gains))
#gvec[0::2]=1.
runs = 1
'''
runs=1
gvec=np.zeros((runs,2*Ndish))
for j in range(runs):
for i in range(len(sim_gains)):
gvec[j,i]=np.random.normal(0,.1,2*Ndish)[i]+sim_gains[i]
#gvec=np.append(rand_gains.real,rand_gains.imag)
'''
#print (gvec,'gvec')
mat = corrcal2.sparse_2level(diag, vecs, src, 2 * lims) #init
'''
mat_ones=mat*np.ones(len(data))
mycov=mat.copy() #copy, init
app_gains=mycov.apply_gains_to_mat(gvec,ant1,ant2) #nothing
#print (np.linalg.pinv(mycov*np.ones((len(data)/2,2))),'my own inv')
mycov_inv=mycov.inv() #inv, copy, init
#print (mycov_inv*np.ones((len(data),len(data))),'mycov_inv')
sd=mycov_inv*data #ans because you're multiplying
chisq=np.sum(sd*data) #nothing
nn=gvec.size/2 #nothing
#print (chisq)
normfac=1.
chisq=chisq+normfac*( (np.sum(gvec[1::2]))**2 + (np.sum(gvec[0::2])-nn)**2) #nothing
print (chisq)
v[edges[i]:edges[i + 1]])
print edges[i], edges[i + 1], mystd
v1 = numpy.zeros(2 * vis.size)
v1[0::2] = 1
v2 = numpy.zeros(2 * vis.size)
v2[1::2] = 1
vecs = numpy.vstack([v1, v2])
src = v1 * 10
big_noise = numpy.zeros(2 * noise.size)
big_noise[0::2] = noise
big_noise[1::2] = noise
big_vis = numpy.zeros(2 * vis.size)
big_vis[0::2] = numpy.real(vis)
big_vis[1::2] = numpy.imag(vis)
mycov = corrcal2.sparse_2level(big_noise, 100 * vecs, 500 * src, 2 * edges)
guess = numpy.zeros(2 * len(ant1))
guess[0::2] = 1.0
fac = 1000.0
from scipy.optimize import fmin_cg
gvec = numpy.zeros(2 * ant1.max() + 2)
gvec[0::2] = 1.0
gvec = gvec + 0.1 * numpy.random.randn(gvec.size)
gvec[0] = 1
gvec[1] = 0
asdf = fmin_cg(corrcal2.get_chisq, gvec * fac, corrcal2.get_gradient,
(big_vis + 500 * src, mycov, ant1, ant2, fac))
for i in range(len(edges)-1):
mystd=numpy.std(u[edges[i]:edges[i+1]])+numpy.std(v[edges[i]:edges[i+1]])
print edges[i],edges[i+1],mystd
v1=numpy.zeros(2*vis.size)
v1[0::2]=1
v2=numpy.zeros(2*vis.size)
v2[1::2]=1
vecs=numpy.vstack([v1,v2])
src=v1*10
big_noise=numpy.zeros(2*noise.size)
big_noise[0::2]=noise
big_noise[1::2]=noise
big_vis=numpy.zeros(2*vis.size)
big_vis[0::2]=numpy.real(vis)
big_vis[1::2]=numpy.imag(vis)
mycov=corrcal2.sparse_2level(big_noise,100*vecs,500*src,2*edges)
guess=numpy.zeros(2*len(ant1))
guess[0::2]=1.0
fac=1000.0
from scipy.optimize import fmin_cg
gvec=numpy.zeros(2*ant1.max()+2)
gvec[0::2]=1.0
gvec=gvec+0.1*numpy.random.randn(gvec.size)
gvec[0]=1
gvec[1]=0
asdf=fmin_cg(corrcal2.get_chisq,gvec*fac,corrcal2.get_gradient,(big_vis+500*src,mycov,ant1,ant2,fac))
isinv=(numpy.fromfile(f,'int32',1)[0]!=0);
nsrc=numpy.fromfile(f,'int32',1);
nblock=numpy.fromfile(f,'int32',1);
nvec=numpy.fromfile(f,'int32',1);
lims=numpy.fromfile(f,'int32',(nblock+1))
diag=numpy.fromfile(f,'float64',n)
vecs=numpy.fromfile(f,'float64',nvec*n)
src=numpy.fromfile(f,'float64',nsrc*n)
n=n[0]
nsrc=nsrc[0]
nvec=nvec[0]
vecs=vecs.reshape([nvec,n])
if nsrc>0:
src=src.reshape([nsrc,n])
mat=corrcal2.sparse_2level(diag,vecs,src,lims,isinv)
f=open('ant1.dat');ant1=numpy.fromfile(f,'int64')-1;f.close()
f=open('ant2.dat');ant2=numpy.fromfile(f,'int64')-1;f.close()
f=open('gtmp.dat');gvec=numpy.fromfile(f,'float64');f.close()
f=open('vis.dat');data=numpy.fromfile(f,'float64');f.close()
niter=0;
t_min=1e-4
t1=time.time()
while (time.time()-t1)<t_min:
grad=corrcal2.get_gradient(gvec,data,mat,ant1,ant2)
niter=niter+1
t2=time.time()
nant=gvec.size/2
#time per gradient
print 'average time was ' + repr((time.time()-t1)/niter)