def forward(self, input, normalise=False):
half0 = self.gatelist[0].forward(input)
half1 = self.gatelist[1].forward(input)
tmp = unionlayer.join(half0, half1)
tmp = unionlayer.join(tmp, self.ancilaries)
for x in range(2, len(self.gatelist)):
tmp = self.gatelist[x].forward(tmp)
return tmp
def forward(self, input):
quarter0 = self.gatelist[0].forward(input)
quarter1 = self.gatelist[1].forward(input)
quarter2 = self.gatelist[2].forward(input)
quarter3 = self.gatelist[3].forward(input)
half0 = unionlayer.join(quarter0, quarter1)
half1 = unionlayer.join(quarter2, quarter3)
tmp = super().forward_two_inputs(half0, half1, 4)
return tmp
def forward(self, input, recursion=0):
third0 = self.gatelist[0].forward(input)
third1 = self.gatelist[1].forward(input)
third2 = self.gatelist[2].forward(input)
tmp = unionlayer.join(third0, third1)
tmp = unionlayer.join(tmp, third2)
for i in range(recursion):
for x in range(3, len(self.gatelist)):
tmp = self.gatelist[x].forward(tmp)
return tmp
def forward_two_inputs(self, input0, input1, start=0):
half0 = self.gatelist[start].forward(input0)
half1 = self.gatelist[start + 1].forward(input1)
tmp = unionlayer.join(half0, half1)
for x in range(start + 2, len(self.gatelist)):
tmp = self.gatelist[x].forward(tmp)
return tmp
def forward(self, input, recursion=1):
half0 = self.gatelist[0].forward(input)
half1 = self.gatelist[1].forward(input)
tmp = unionlayer.join(half0, half1)
for i in range(recursion):
for x in range(2, len(self.gatelist)):
tmp = self.gatelist[x].forward(tmp)
return tmp
def fidelity_partial_list(output, target,nbqubitout,nbqutbittarget):
#print(output)
#print(target)
partial=tf.ones([2**(nbqubitout-nbqutbittarget),tf.shape(output)[1]],dtype=output.dtype)
#partialnorm=tf.div(partial, tf.norm(partial, axis=0))
target2=unionlayer.join(partial,target)
#print(target2)
temp=tf.abs(tf.reduce_sum(tf.square(tf.abs(tf.einsum('ij,ij->ij', tf.conj(target2), output))), axis=0))
#print(temp)
return temp
def preprocess(input,target):
#ctes=vectorencoder.gennormpluscte(castinput,3,cte=0.01)
#ancinit = 3 * tf.ones_like(target)
#ancinit2 = 63*tf.ones_like(target)
#anc0 = unionlayer.inttoqubit(ancinit, 8)
#anc1 = unionlayer.inttoqubit(ancinit2, 6)
ctes = vectorencoder.gencte(input, 2, cte=0.0)
#print(ctes)
encodeddata=vectorencoder.encode_vectors(input,2,ctes)
colvec =tf.transpose(encodeddata)
#print(colvec.shape)
c = unionlayer.join(colvec, colvec)
#v = unionlayer.join(c, c)
#w = unionlayer.join(v, v)
#w = unionlayer.join(v, anc0 )
#w = unionlayer.join(anc1, w)
#w = unionlayer.join(c ,anc0 )
return c
init = tf.orthogonal_initializer
#real = tf.get_variable("v", [2**8, 2**8], initializer=init)
real = tf.eye(2**8, dtype="float32")
imag = tf.zeros_like(real, dtype="float32")
param = tf.complex(real, imag)
learningrate = 0.01
momentum = 0.01
gate0 = genericQGate.genericQGate(param, 8, 8, 0, learningrate, momentum)
gate1 = genericQGate.genericQGate(param, 8, 8, 0, learningrate, momentum)
gate2 = genericQGate.genericQGate(param, 8, 8, 0, learningrate, momentum)
copy0 = unionlayer.join(normedinput, normedinput)
copy1 = unionlayer.join(copy0, normedinput)
full = unionlayer.join(
copy1,
unionlayer.inttoqubit(tf.zeros_like(targetbatch, dtype="float32"), 2))
o0 = gate0.forward(full)
for x in range(0, 6):
o0, r = lazymeasure.lazymeasure(o0)
full1 = unionlayer.join(copy1, o0)
o1 = gate1.forward(full1)
for x in range(0, 6):
o1, r = lazymeasure.lazymeasure(o1)
label=[[1],[0]]
#print(fidelity_partial_list(un,zero,2,1))
fd_list=fidelity_partial_list(un, zero, 2, 1)
print(fd_list)
#print(true_max_metric(fd_list))
# print(cross_entropy(un, zero,2,1))
# label=tf.transpose(tf.cast([0.0,1.0], dtype="float32"))
#print(maxclass(un, label, 2, 1, 2, 'inttoqubit'))
#print(true_max_metric(un,label,2,1,'inttoqubit'))
real = [[1 / tf.sqrt(2.0), 1 / tf.sqrt(2.0)],[1 / tf.sqrt(2.0), 1 / tf.sqrt(2.0)]]
imag = [[0.0, 0.0],[0.0, 0.0]]
plus = tf.transpose(tf.complex(real, imag))
temp=unionlayer.join(plus,plus)
testtarget0=unionlayer.inttoqubit([0.0,0.0], 1)
testtarget1 = unionlayer.inttoqubit([1.0, 1.0], 1)
tep=tf.random.uniform([256,2])
test3=tf.transpose(unionlayer.state_activation_0(tf.transpose(tep)))
print(test3)
print(np.linalg.norm(test3, ord=2, axis=0))
test4=tf.transpose(tf.nn.softmax(tf.transpose(tep)))
print(test4)
print(np.sum(test4,axis=0))
#print(test3)
#print(test)
#print(y)
#print(lf.shape)
#print(l1)
#print(lf)
#print(tf.norm(lf,axis=1))
#tmp=1 /tf.norm(lf, axis=1)
#tmp2=tf.einsum('ij,i->ij',lf,tmp)
#print("lol")
#print(tmp2)
#lfn=tf.div(lf,tf.norm(lf,2,axis =1))
#print(lfn.shape)
#print(tf.norm(tmp2, axis=1))
nninput = tf.transpose(lf)
#sth=unionlayer.join(anc0,nninput )
#st=unionlayer.join(sth,anc1)
st = unionlayer.join(anc0, nninput)
#cir = subcircuit.ArityFillCircuit(nbqubits, 8, 6, "test0", learningrate, momentum)
#cir = subcircuit.ArityFillCircuit(nbqubits, aritycircuitsize, aritycircuitdepth, "test0", learningrate, momentum)
#cir = subcircuit.ArityHalfCircuit(nbqubits, aritycircuitsize, aritycircuitdepth, "test0", learningrate, momentum)
cir = subcircuit.QuincunxCircuit(nbqubits, aritycircuitsize,
aritycircuitdepth, "test0")
#out = cir.forward_two_inputs(nninput,stateinput)
#out = cir.forward_two_inputs(nninput, nninput)
#out = cir.forward(nninput)
out = cir.forward(st, 1)
#for x in range(0,8):
gate2 = genericQGate.genericQGate(phasechange, 1, 1, 0, learningrate,
momentum)
#for i in range(8):
real = tf.random_normal([2, batchsize - 2], 0, 1)
imag = tf.random_normal([2, batchsize - 2], 0, 1)
#real = [[0.0, 1.0]]
#imag = [[0.0, 0.0]]
temp = tf.complex(real, imag)
vectorrand = tf.div(temp, tf.norm(temp, axis=0))
vectorrand = tf.concat([tf.transpose(param1), vectorrand], axis=1)
prep = unionlayer.join(vectorrand, phipluscte)
#o0 = gateh.forward(gate0.forward(prep))
o0 = gatep1.forward(gatep.forward(prep))
(o1zero, r0zero), (o1un, r0un) = lazymeasure.lazymeasureboth(o0)
(o2zz, r1zz), (o2zu, r1zu) = lazymeasure.lazymeasureboth(o1zero)
(o2uz, r1uz), (o2uu, r1uu) = lazymeasure.lazymeasureboth(o1un)
o2 = tf.concat(
[o2zz,
gate1.forward(o2zu), o2uz,
gate2.forward(gate1.forward(o2uu))],
axis=1)