# Load the inferred values:
a_new = np.loadtxt('anew.txt')
#print a_new
j_old = 0.01
count = 0
Nq = 2000
## Decimation algorithm:
## Null couplings:
J0 = a_new[np.where(a_new <= -0.03)]
#Decimate the small values:
J0s = a_new[np.where(a_new <= -0.03)] = 0.0
n1 = (a_new == 0).sum() # count the zero values
print n1
# Non-zero couplings:
J2 = a_new[np.where(a_new >= -0.03)]
j2s = np.count_nonzero(a_new)
#print J2,j2s
for i in range(500):
    j0s = 0.0
    #maximize:
    x = np.random.uniform(0.2,0.4)
    Lmax = np.log(1-x)*(s2[j1])-((1-x)*sum1/1-np.power((1-x), sum1))*(np.log(1-x)*s2[j1])
    #print Lmax
    #infer the couplings:
    jnew = j_old + (rate*Lmax[:,44])
    #sort inferred couplings by absolute values:
    jabs = np.absolute(jnew)
    #print jabs
    j1s = jabs[np.where(jabs < 0.0009)] = 0.0
    #print jabs
    #move the least couplings from j1s to j0s:
    #
    jabs1 = np.extract(j1s == 0.0, j1s) #j1s ==0.0
    jabs2 = np.extract(jabs >= 0.0009, jabs)
    #print jabs2
    if(np.any(jabs != 0.0)):
       count = count+1
       print count, jabs, j1s
       #Nq = (100*99)/2
       #x1 = (Nq - jj)/ Nq # jj: fraction of non-decimated couplings:
       #x1 = np.arange(1,0,-0.01)
       x1  = (Nq-j1s)/Nq
       xs = (Nq-count)/2 # of decimated couplings:
       #print x1
       #define the tilted likelihood:
       tplf = L1[:,44]-x1*Lmax[:,44]-(1-x1)*(-(np.log(2)))
       #print tplf, x1, Nq, j1s
       #print tplf, xs, x1