def initial_ipeps(self):
dim = (self.d, ) + (self.chi, ) * self.nVirtual
if self.iniWay is 'random':
tensor = tm.symmetrical_rand_peps_tensor(self.d, self.chi,
self.nVirtual)
if self.stateType is 'mixed':
bond = (self.d0, self.d0) + (self.chi, ) * self.nVirtual
tensor = tensor.reshape(bond)
ind = (1, 0) + tuple(range(2, self.nVirtual + 2))
tensor = (tensor + tensor.transpose(ind)) / 2
bond = (self.d, ) + (self.chi, ) * self.nVirtual
tensor = tensor.reshape(bond)
for n in range(0, self.nTensor):
self.tensors[n] = tensor.copy()
elif self.iniWay is 'ones':
for n in range(0, self.nTensor):
self.tensors[n] = np.ones(dim)
elif self.iniWay is 'id':
if self.stateType is 'mixed':
if self._is_debug:
if abs(self.d0**2 - self.d) > 1e-10:
bf.print_error('For mixed state, d should be as d0^2. '
'Check self.d or self.stateType')
for n in range(0, self.nTensor):
self.tensors[n] = np.eye(
self.d0).reshape((self.d, ) + (1, ) * self.nVirtual)
else:
bf.print_error('Initial way "id" is only for thermal states')
def decision_mps(para=None):
if para is None:
para = pm.parameters_decision_mps()
a = TNmachineLearning.DecisionTensorNetwork(
para['dataset'],
2,
para['chi'],
'mps',
para['classes'],
para['numbers'],
if_reducing_samples=para['if_reducing_samples'])
a.images2vecs_test_samples(para['classes'])
print(a.vLabel)
for n in range(0, a.length):
bf.print_sep()
print('Calculating the %i-th tensor' % n)
a.update_tensor_decision_mps_svd(n)
# a.update_tensor_decision_mps_svd_threshold_algo(n)
# a.update_tensor_decision_mps_gradient_algo(n)
a.update_v_ctr_train(n)
a.calculate_intermediate_accuracy_train(n)
if a.remaining_samples_train.__len__() == 0:
print('All samples are classified correctly. Training stopped.')
break
print('The current accuracy = %g' % a.intermediate_accuracy_train[n])
print('Entanglement: ' + str(a.lm[n].reshape(1, -1)))
if para['if_reducing_samples']:
print('Number of remaining samples: ' +
str(a.remaining_samples_train.__len__()))
def show_image(self, n):
# n can be a number or an image
if type(n) is int:
l_side = int(np.sqrt(self.length))
bf.plot_surf(np.arange(0, l_side), np.arange(0, l_side),
self.images[:, n].reshape(l_side, l_side))
else:
l_side = n.shape[0]
bf.plot_surf(np.arange(0, l_side), np.arange(0, l_side), n)
def __init__(self,
lattice,
chi,
state_type='pure',
spin='half',
ini_way='random',
operators=None,
is_symme_env=True,
is_debug=False):
PepsBasic.__init__(self)
"""
Order of bonds: (phys, virtual, virtual, ...)
Lattices:
'honeycomb0' lattice: n-th virtual bond, n-th lambda
0,lm0 1,lm1
\ /
- 2,lm2 -
/ \
1,lm1 0,lm0
"""
self.lattice = lattice
self.stateType = state_type # pure or mixed
self.spin = spin
self.d0 = from_spin2phys_dim(spin)
self.chi = chi
self.nTensor = 0
self.nLm = 0
self.nVirtual = 0
self.pos_lm = list()
self.lm_ten_bond = None
self.iniWay = ini_way
self._is_debug = is_debug
if self.stateType is 'mixed':
self.d = self.d0**2
else:
self.d = self.d0
self.initial_lattice()
self.tensors = bf.empty_list(self.nTensor)
self.lm = bf.empty_list(self.nLm)
self.initial_ipeps()
self.initial_lm()
if operators is None:
op_half = spin_operators(spin)
self.operators = [
op_half['id'], op_half['sx'], op_half['sy'], op_half['sz'],
op_half['su'], op_half['sd']
]
else:
self.operators = operators
# Below is for tree DMRG
self.is_symme_env = is_symme_env
self.env = None
self.model_related = dict()
def check_consistency(self):
if self.dataInfo['NumTotalTrain'] != sum(self.dataInfo['nClassNum']):
bf.print_error('The total number in the dataset is NOT consistent '
'with the sum of the samples of all classes')
for n in range(0, self.dataInfo['NumClass']):
start = self.dataInfo['nStart'][n]
end = start + self.dataInfo['nClassNum'][n]
tmp = self.labels[start:end]
if not np.prod(tmp == tmp[0]):
bf.print_error('In the ' + str(tmp[0]) + '-th labels, not all labels are '
+ str(tmp[0]))
print(bf.arg_find_array(tmp != tmp[0]))
def rho_one_body_simple(self, which_t):
if which_t is 'all':
rho = bf.empty_list(self.nTensor)
for nt in range(0, self.nTensor):
rho[nt] = self.rho_one_body_simple_nt_tensor(nt)
elif type(which_t) is int:
rho = self.rho_one_body_simple_nt_tensor(which_t)
else:
rho = bf.empty_list(len(which_t))
n = 0
for nt in which_t:
rho[n] = self.rho_one_body_simple_nt_tensor(nt)
n += 1
return rho
def rho_two_body_simple(self, which_lm):
if which_lm is 'all':
rho = bf.empty_list(self.nLm)
for n_lm in range(0, self.nLm):
rho[n_lm] = self.rho_two_body_nlm_simple(n_lm)
elif type(which_lm) is int:
rho = self.rho_two_body_nlm_simple(which_lm)
else:
rho = bf.empty_list(len(which_lm))
n = 0
for n_lm in which_lm:
rho[n] = self.rho_two_body_nlm_simple(n_lm)
n += 1
return rho
def initial_lm(self):
lm = np.random.rand(self.chi, )
lm /= np.linalg.norm(lm)
if self.iniWay is 'random':
for n in range(0, self.nLm):
self.lm[n] = lm.copy()
elif self.iniWay is 'ones':
for n in range(0, self.nLm):
self.tensors[n] = np.ones((self.chi, )) / (self.chi**0.5)
elif self.iniWay is 'id':
if self.stateType is 'mixed':
for n in range(0, self.nLm):
self.lm[n] = np.ones(1, )
else:
bf.print_error('Initial way "id" is only for thermal states')
def gtnc(para_tot=None):
print('Preparing parameters')
if para_tot is None:
para_tot = pm.parameters_gcmpm()
n_class = len(para_tot['classes'])
paras = bf.empty_list(n_class)
for n in range(0, n_class):
paras[n] = copy.deepcopy(para_tot)
paras[n]['class'] = int(para_tot['classes'][n])
paras[n]['chi'] = para_tot['chi'][n]
paras[n]['theta'] = para_tot['theta']
paras[n]['save_exp'] = save_exp_gtn_one_class(paras[n])
classifiers = bf.empty_list(n_class)
for n in range(0, n_class):
print_dict(paras[n])
data = para_tot['data_path'] + paras[n]['save_exp']
if para_tot['if_load'] and os.path.isfile(data):
print('The classifier already exists. Load directly')
classifiers[n] = load_pr(data, 'a')
else:
print('Training the MPS of ' + str(para_tot['classes'][n]))
classifiers[n] = gtn_one_class(paras[n])[0]
# if para_tot['if_save']:
# save_pr('../data_tnml/gcmpm/', paras[n]['save_exp'],
# [classifiers[n]], ['classifier'])
# classifiers[n].mps.check_orthogonality_by_tensors(tol=1e-12)
# ==================== Testing accuracy ====================
print('Calculating the testing accuracy')
b = TNmachineLearning.MachineLearningFeatureMap(para_tot['d'])
b.load_data(data_path='..\\..\\..\\MNIST\\',
file_sample='t10k-images.idx3-ubyte',
file_label='t10k-labels.idx1-ubyte',
is_normalize=True)
b.select_samples(para_tot['classes'])
if classifiers[0].is_dct:
b.dct(shift=para_tot['shift'], factor=para_tot['factor'])
b.images2vecs(para_tot['theta'] * np.pi / 2)
fid = bf.empty_list(n_class)
for n in range(0, n_class):
fid[n] = b.compute_fidelities(classifiers[n].mps.mps)
max_fid = np.argmin(np.hstack(fid), axis=1)
predict = np.zeros(max_fid.shape, dtype=int)
for n in range(0, n_class):
predict += (max_fid == n) * int(para_tot['classes'][n])
# plot(predict)
# plot(b.labels)
accuracy = np.sum(predict == b.labels, dtype=float) / b.numVecSample
print(accuracy)
def find_pos_of_lm(self):
self.lm_ten_bond = np.zeros((self.nLm, 2, 2), dtype=int)
for n_lm in range(0, self.nLm):
n_found = 0
for n in range(0, self.nTensor):
if n_lm in self.pos_lm[n]:
self.lm_ten_bond[n_lm, n_found, 0] = n
self.lm_ten_bond[n_lm, n_found,
1] = self.pos_lm[n].index(n_lm)
n_found += 1
if n_found == 2:
break
if self._is_debug and n_found < 2:
bf.print_error(
'In "find_pos_of_one_lm", n_found is ony %g. It should 2' %
n_found)
def initialize_env(self):
self.env = bf.empty_list(self.nVirtual)
tmp = np.random.randn(self.chi, self.d0**2, self.chi)
tmp = (tmp + tmp.transpose([2, 1, 0])) / 2
tmp /= np.linalg.norm(tmp.reshape(-1, ))
for n in range(0, self.nVirtual):
self.env[n] = tmp.copy()
def read_mps(load,):
load_path = load['path']
load_exp = load['exp']
if load['Is_continue']:
data = bf.load_pr(load_path+load_exp+'_.pr')
else:
data = bf.load_pr(load_path+load_exp+'.pr')
print(load_path+load_exp+'.pr')
A = data['A']
n = len(A.mps)
dims=[1]
for i in range(n-1):
dims.append(A.mps[i].shape[2])
dims.append(1)
return A.mps, dims
def __init__(self, data, parity=0):
self.parity = parity
if type(data) is dict:
self.data = data
one_data = data['00']
self.ndim = one_data.ndim
self.shape = [one_data.shape[n] * 2 for n in range(0, self.ndim)]
else:
self.ndim = data.ndim
self.shape = list(data.shape)
self.data = dict()
self.normal_to_z2(data)
for n in range(0, self.ndim):
if self.shape[n] % 2 == 1:
bf.print_error(
'DimError: for a Z2 tensor, all bond dimensions should be even'
)
def one_bond_so_transformation(self, nt1, vb1, nt2, vb2):
# Super-orthogonal transformation on one virtual bond
# vb does NOT count the physical bond
if self._is_debug:
if self.pos_lm[nt1][vb1] != self.pos_lm[nt2][vb2]:
bf.print_error(
'In one_bond_so_transformation, the two virtual bonds must'
'correspond to the same lambda')
m1 = self.bond_env_matrix_simple(nt1, vb1)
m2 = self.bond_env_matrix_simple(nt2, vb2)
flag = False
if self._is_debug:
_lm = self.lm[self.pos_lm[nt1][vb1]].copy()
flag = (self.chi == self.tensors[nt1].shape[vb1 + 1])
u1, u2, self.lm[self.pos_lm[nt1]
[vb1]] = tm.transformation_from_env_mats(
m1,
m2,
self.lm[self.pos_lm[nt1][vb1]],
self.chi,
norm_way=1)[:3]
if self._is_debug and flag:
_tmp = u1.dot(np.diag(self.lm[self.pos_lm[nt1][vb1]])).dot(u2.T)
err = np.linalg.norm(tm.off_diagonal_mat(_tmp).reshape(-1, ))
if err > 1e-10:
print(
'Warning of the transformations from environment: not diagonal (%g)'
% err)
_tmp = np.diag(_tmp)
_tmp = _tmp / np.linalg.norm(_tmp)
err = np.linalg.norm(_tmp - self.lm[self.pos_lm[nt1][vb1]])
if err > 1e-10:
print(
'Warning of the transformations from environment: not recover lm (%g)'
% err)
print(self.lm[self.pos_lm[nt1][vb1]])
self.tensors[nt1] = tm.absorb_matrix2tensor(self.tensors[nt1], u1,
vb1 + 1)
self.tensors[nt2] = tm.absorb_matrix2tensor(self.tensors[nt2], u2,
vb2 + 1)
self.tensors[nt1] /= max(abs(self.tensors[nt1].reshape(-1, 1)))
self.tensors[nt2] /= max(abs(self.tensors[nt2].reshape(-1, 1)))
# self.lm[self.pos_lm[nt1][vb1]] = tm.normalize_tensor(self.lm[self.pos_lm[nt1][vb1]])[0]
return m1, m2
def images2vecs(self, classes, numbers, how='random'):
self.vec_classes = classes
num_class = classes.__len__()
ntot = 0
if numbers is None:
numbers = ['all'] * num_class
for n in range(0, num_class):
if numbers[n] is 'all':
ntot += self.dataInfo['nClassNum'][n]
else:
ntot += min(numbers[n], self.dataInfo['nClassNum'][n])
self.numVecSample = ntot
self.LabelNow = np.zeros((ntot,), dtype=int)
self.tmp = np.zeros((self.length, ntot))
n_now = 0
for n in range(0, num_class):
if numbers[n] is 'all':
start = self.dataInfo['nStart'][n]
end = start + self.dataInfo['nClassNum'][n]
self.tmp[:, n_now:self.dataInfo['nClassNum'][n]] = self.images[:, start:end]
self.LabelNow[n_now:self.dataInfo['nClassNum'][n]] = self.labels[start:end]
n_now += self.dataInfo['nClassNum'][n]
else:
n_sample = numbers[n]
start = self.dataInfo['nStart'][classes[n]]
if n_sample >= self.dataInfo['nClassNum'][classes[n]]:
rand_p = range(start, self.dataInfo['nClassNum'][classes[n]] + start)
elif how is 'random':
rand_p = np.random.permutation(self.dataInfo['nClassNum'][classes[n]])[
:n_sample] + start
elif how is 'first':
rand_p = range(start, n_sample + start)
else:
rand_p = range(self.dataInfo['nClassNum'][classes[n]] - n_sample + start,
self.dataInfo['nClassNum'][classes[n]] + start)
for ns in rand_p:
self.tmp[:, n_now] = self.images[:, ns]
self.LabelNow[n_now] = self.labels[ns]
n_now += 1
self.multiple_images2vecs()
self.clear_tmp_data()
if is_debug and n_now != ntot:
bf.print_error('In images2vecs_train_samples: total number of vectorized '
'images NOT consistent')
def initial_lattice(self):
# pos_lm[nt][nb]=x denotes the x-th lm locates at the nb-th bond of the nt-th tensor
if self.lattice is 'honeycomb0':
self.nTensor = 2
self.nLm = 3
self.nVirtual = 3
# means for the 0-the tensor, it is the 0, 1, and 2-th lm on the three virtual bonds
self.pos_lm = [[], []]
self.pos_lm[0] = [0, 1, 2]
self.pos_lm[1] = [0, 1, 2]
elif self.lattice is 'honeycombTreeDMRG':
# for tree DMRG of honeycomb lattice; the TN is square (one tensor with two spins)
self.nTensor = 5
self.nVirtual = 4
self.d = self.d0**2
else:
bf.print_error('Incorrect input of the lattice')
self.find_pos_of_lm()
def __init__(self, dataset='mnist', data_path='..\\..\\..\\MNIST\\',
file_sample='train-images.idx3-ubyte', file_label='train-labels.idx1-ubyte',
is_normalize=True, par_pool=None):
# MNIST files for training: 'train-images.idx3-ubyte', 'train-labels.idx1-ubyte'
# MNIST files for testing: 't10k-images.idx3-ubyte', 't10k-labels.idx1-ubyte'
self.dataset = dataset
self.dataInfo = dict()
self.images = bf.decode_idx3_ubyte(path.join(data_path, file_sample))
self.labels = bf.decode_idx1_ubyte(path.join(data_path, file_label))
self.length = self.images.shape[0]
if is_normalize:
self.images /= 256
self.analyse_dataset()
self.tmp = None
self.nPool = par_pool
self.parPool = None
if is_debug:
self.check_consistency()
def get_model_related_tree_dmrg(self, jx, jy, jz, hx, hz, tau=1e-6):
if 'h2phys' not in self.model_related:
self.model_related['h2phys'] = hamiltonian_heisenberg(
self.spin, jx, jy, jz, hx, hz)
self.model_related['h2_gate'] = expm(-tau / 2 *
self.model_related['h2phys'])
if 'tensor_gate' not in self.model_related:
self.model_related['tensor_gate'] = hamiltonian2gate_tensors(
self.model_related['h2phys'], tau)
if 'hbath' not in self.model_related:
self.model_related['hbath'] = bf.empty_list(self.nVirtual)
def load_data(self,
data_path=None,
file_sample=None,
file_label=None,
is_normalize=None):
# MNIST files for training: 'train-images.idx3-ubyte', 'train-labels.idx1-ubyte'
# MNIST files for testing: 't10k-images.idx3-ubyte', 't10k-labels.idx1-ubyte'
if data_path is None:
data_path = self.data_path
if file_label is not None:
self.data_samples = path.join(data_path, file_sample)
if self.is_there_labels:
self.data_labels = path.join(data_path, file_label)
if not (is_normalize is None):
self.is_normalize_data = is_normalize
self.images = bf.decode_idx3_ubyte(self.data_samples)
if self.is_there_labels:
self.labels = bf.decode_idx1_ubyte(self.data_labels)
self.length = self.images.shape[0]
if self.is_normalize_data:
self.images /= (254 * (self.images.max() > 2) + 1)
def gcmpm_one_class(para=None):
if para is None:
para = pm.parameters_gcmpm_one_class()
para['save_exp'] = save_exp_gcmpm_one_class(para)
if para['parallel'] is True:
par_pool = para['n_nodes']
else:
par_pool = None
if para['if_load'] and os.path.isfile(para['save_exp']):
a = bf.load_pr(os.path.join(para['data_path'], para['save_exp']), 'a')
else:
a = TNmachineLearning.MachineLearningMPS(para['d'], para['chi'], para['dataset'],
par_pool=par_pool)
a.images2vecs([para['class']], [100])
a.initialize_virtual_vecs_train()
a.update_virtual_vecs_train('all', 'all', 'both')
a.mps.correct_orthogonal_center(0, normalize=True)
a.mps.mps[0] /= np.linalg.norm(a.mps.mps[0].reshape(-1, ))
mps0 = a.mps.mps.copy()
for t in range(0, para['sweep_time']):
# from left to right
if para['if_print_detail']:
print('At the ' + str(t) + '-th sweep, from left to right')
for nt in range(0, a.length):
a.update_tensor_gradient(nt, para['step'])
if nt != a.length-1:
a.update_virtual_vecs_train('all', nt, 'left')
# from left to right
print('At the ' + str(t) + '-th sweep, from right to left')
for nt in range(a.length-1, -1, -1):
a.update_tensor_gradient(nt, para['step'])
if nt != 0:
a.update_virtual_vecs_train('all', nt, 'right')
if t > para['check_time0'] and ((t+1) % para['check_time'] == 0
or t+1 == para['sweep_time']):
fid = ln_fidelity_per_site(mps0, a.mps.mps)
if fid < (para['step'] * para['ratio_step_tol']):
print('After ' + str(t+1) + ' sweeps: fid = %g' % fid)
para['step'] *= para['step_ratio']
elif t+1 == para['sweep_time']:
print('After all ' + str(t+1) + ' sweeps finished, fid = %g. '
'Consider to increase the sweep times.' % fid)
else:
print('After ' + str(t+1) + ' sweeps, fid = %g.' % fid)
mps0 = a.mps.mps.copy()
if para['step'] < para['step_min']:
print('Now step = ' + str(para['step']) + ' is sufficiently small. Break the loop')
break
else:
print('Now step = ' + str(para['step']))
if para['if_save']:
save_pr(para['data_path'], para['save_exp'], [a, para], ['a', 'para'])
return a, para
def update_by_given_effective_ops(psi, ops, bonds):
indexes = bf.empty_list(1 + bonds.__len__())
indexes[0] = list(range(psi.ndim))
x = 1
for n in range(psi.ndim):
if n in bonds:
indexes[0][n] = x
indexes[bonds.index(n) + 1] = [-n - 1, x]
x += 1
else:
indexes[0][n] = -n - 1
return tm.cont([psi] + ops, indexes)
def show_image(self, n, title_way='which_class'):
# n can be a number or an image
if type(n) is int:
l_side = int(np.sqrt(self.length))
bf.show_multiple_images_v1(
[self.images[:, n].reshape(l_side, l_side)], titles=[n])
elif type(n) is np.ndarray:
bf.show_multiple_images_v1([n])
else:
l_side = int(np.sqrt(self.length))
num = n.__len__()
n_now = 0
while num > 0.5:
dn = min(num, 100)
if title_way is 'which_sample':
titles = [x for x in n[n_now:n_now + dn]]
elif title_way is 'which_class':
titles = [int(self.labels[x]) for x in n[n_now:n_now + dn]]
else:
titles = None
# bf.show_multiple_images_v0(
# [self.images[:, x].reshape(l_side, l_side) for x in n[n_now:n_now + dn]],
# gap=0)
bf.show_multiple_images_v1([
self.images[:, x].reshape(l_side, l_side)
for x in n[n_now:n_now + dn]
],
titles=titles)
n_now += dn
num -= dn
def gcmpm(para_tot=None):
print('Preparing parameters')
if para_tot is None:
para_tot = pm.parameters_gcmpm()
n_class = len(para_tot['classes'])
paras = bf.empty_list(n_class)
for n in range(0, n_class):
paras[n] = para_tot.copy()
paras[n]['class'] = para_tot['classes'][n]
paras[n]['chi'] = para_tot['chi'][n]
paras[n]['save_exp'] = save_exp_gcmpm_one_class(paras[n])
classifiers = bf.empty_list(n_class)
for n in range(0, n_class):
data = '../data_tnml/gcmpm/' + paras[n]['save_exp']
if para_tot['if_load'] and os.path.isfile(data):
print('The classifier already exists. Load directly')
classifiers[n] = load_pr(data, 'classifier')
else:
print('Training the MPS of ' + str(para_tot['classes'][n]))
classifiers[n] = gcmpm_one_class(paras[n])[0]
if para_tot['if_save']:
save_pr('../data_tnml/gcmpm/', paras[n]['save_exp'],
[classifiers[n]], ['classifier'])
# Testing accuracy
print('Calculating the testing accuracy')
labels = para_tot['classes']
b = TNmachineLearning.MachineLearningFeatureMap('MNIST', para_tot['d'],
file_sample='t10k-images.idx3-ubyte',
file_label='t10k-labels.idx1-ubyte')
b.images2vecs(para_tot['classes'], ['all', 'all'])
fid = np.zeros((n_class, ))
num_wrong = 0
for ni in range(0, b.numVecSample):
for n in range(0, n_class):
fid[n] = b.fidelity_mps_image(classifiers[n].mps.mps, ni)
n_max = int(np.argmax(fid))
if labels[n_max] != b.LabelNow[ni]:
num_wrong += 1
accuracy = num_wrong/b.numVecSample
print(accuracy)
def initialize_virtual_vecs_train(self, way='contract'):
self.norms = np.ones((self.length, self.numVecSample))
self.vecsLeft = bf.empty_list(self.length)
self.vecsRight = bf.empty_list(self.length)
if way is 'random':
for n in range(0, self.length):
self.vecsLeft[n] = np.random.randn(self.mps.virtual_dim[n],
self.numVecSample)
self.vecsRight[n] = np.random.randn(
self.mps.virtual_dim[n + 1], self.numVecSample)
elif way is 'ones':
for n in range(0, self.length):
self.vecsLeft[n] = np.ones(
(self.mps.virtual_dim[n], self.numVecSample))
self.vecsRight[n] = np.ones(
(self.mps.virtual_dim[n + 1], self.numVecSample))
else:
self.vecsRight[self.length - 1] = np.ones(
(self.mps.virtual_dim[self.length], self.numVecSample))
self.update_virtual_vecs_train_all_tensors('right')
self.vecsLeft[0] = np.ones(
(self.mps.virtual_dim[0], self.numVecSample))
def initial_lattice(self):
# pos_lm[nt][nb]=x denotes the x-th lm locates at the nb-th bond of the nt-th tensor
if self.lattice is 'honeycomb0':
self.nTensor = 2
self.nLm = 3
self.nVirtual = 3
# self.pos_lm[x] means for the x-th tensor, it is the 0, 1, and 2-th lm on the three virtual bonds
self.pos_lm = [[], []]
self.pos_lm[0] = [0, 1, 2]
self.pos_lm[1] = [0, 1, 2]
elif self.lattice is 'honeycombTreeDMRG':
# for tree DMRG of honeycomb lattice; the TN is square (one tensor with two spins)
self.nTensor = 5
self.nVirtual = 4
self.d = self.d0**2
elif self.lattice in ['kagome', 'husimi']:
self.nTensor = 2 # use symmetrical environment: only one orthogonal tensor
self.nVirtual = 3
self.d = self.d0
self.stateType = 'pure'
else:
bf.print_error('Incorrect input of the lattice')
self.find_pos_of_lm()
def rho_two_body_nlm_simple(self, n_lm):
nt1 = self.lm_ten_bond[n_lm, 0, 0]
vb1 = self.lm_ten_bond[n_lm, 0, 1]
nt2 = self.lm_ten_bond[n_lm, 1, 0]
vb2 = self.lm_ten_bond[n_lm, 1, 1]
if self._is_debug:
if n_lm != self.pos_lm[nt2][vb2]:
bf.print_error(
'In rho_two_body_simple, the two virtual bonds must'
'correspond to the same lambda')
bonds = list(range(0, self.nVirtual))
bonds.remove(vb1)
tmp1 = self.absorb_lm(nt1, False, bonds)
tmp2 = self.absorb_lm(nt2, False, 'all')
if self.stateType is 'pure':
bonds = list(range(1, self.nVirtual + 1))
bonds.remove(vb1 + 1)
tmp1 = np.tensordot(tmp1.conj(), tmp1, (bonds, bonds))
bonds = list(range(1, self.nVirtual + 1))
bonds.remove(vb2 + 1)
tmp2 = np.tensordot(tmp2.conj(), tmp2, (bonds, bonds))
elif self.stateType is 'mixed':
s = tmp1.shape
bonds = list(range(1, self.nVirtual + 2))
bonds.remove(vb1 + 2)
tmp1 = tmp1.reshape((self.d0, self.d0) + s[1:])
tmp1 = np.tensordot(tmp1.conj(), tmp1, (bonds, bonds))
s = tmp2.shape
bonds = list(range(1, self.nVirtual + 2))
bonds.remove(vb2 + 2)
tmp2 = tmp2.reshape((self.d0, self.d0) + s[1:])
tmp2 = np.tensordot(tmp2.conj(), tmp2, (bonds, bonds))
rho = tm.cont([tmp1, tmp2], [[-1, 1, -3, 2], [-2, 1, -4, 2]])
rho = rho.reshape(self.d0 * self.d0, self.d0 * self.d0)
rho = (rho + rho.conj().T) / 2
rho /= np.trace(rho)
return rho
def observe_by_features(self, features, pos):
if len(pos) == self.length:
bf.print_error(
'Input features cannot be as many as the total features')
features = np.array(features).reshape(-1, 1)
features = self.map_to_vectors(features).squeeze()
data_mps = self.mps.wrap_data()
mps = MPS(self.length, self.d, self.chi)
mps.refresh_mps_properties(data_mps)
for n in range(np.array(pos).size):
mps.mps[pos[n]] = np.tensordot(mps.mps[pos[n]], features[:, n],
[[1], [0]])
pos = np.sort(pos)
for n in pos[::-1]:
if n > 0:
mps.mps[n - 1] = np.tensordot(mps.mps[n - 1], mps.mps[n],
[[mps.mps[n - 1].ndim - 1], [0]])
else:
mps.mps[n + 1] = np.tensordot(mps.mps[n], mps.mps[n + 1],
[[1], [0]])
mps.mps.__delitem__(n)
mps.refresh_mps_properties()
mps.correct_orthogonal_center(0, normalize=True)
return mps
def analyse_dataset(self):
# Total number of samples
self.dataInfo['NumTotalTrain'] = self.labels.__len__()
# Order the samples and labels
order = np.argsort(self.labels)
self.images = bf.sort_vecs(self.images, order, which=1)
self.labels = np.array(sorted(self.labels))
# Total number of classes
self.dataInfo['NumClass'] = int(self.labels[-1] + 1)
# Detailed information
self.dataInfo['nStart'] = np.zeros((self.dataInfo['NumClass'], ), dtype=int)
self.dataInfo['nClassNum'] = np.zeros((self.dataInfo['NumClass'],), dtype=int)
self.dataInfo['nStart'][0] = 0
for n in range(1, self.dataInfo['NumClass']):
x = tm.arg_find_array(self.labels[self.dataInfo['nStart'][n-1] + 1:] == n,
1, 'first')
self.dataInfo['nClassNum'][n-1] = x + 1
self.dataInfo['nStart'][n] = x + self.dataInfo['nStart'][n-1] + 1
self.dataInfo['nClassNum'][-1] = \
self.dataInfo['NumTotalTrain'] - self.dataInfo['nStart'][-1]
def __init__(self, dataset='mnist'):
self.dataset = dataset
self.dataInfo = dict()
self.is_normalize_data = True
self.is_there_labels = True # For generative tasks, no labels are needed
self.data_samples = ''
self.data_labels = ''
self.images = np.zeros(0)
self.labels = np.zeros(0, dtype=int)
self.length = -1
self.img_size = [0, 0]
self.is_dct = False
self.dct_info = dict()
self.is_label_added = False
self.project_path = bf.project_path()
self.data_path = None
self.identify_dataset()
self.tmp = None
if is_debug:
self.check_consistency()
def initialize_virtual_vecs_train(self):
self.vecsLeft = bf.empty_list(self.length, list())
self.vecsRight = bf.empty_list(self.length, list())
for n in range(0, self.length):
self.vecsLeft[n] = np.ones((self.mps.virtual_dim[n], self.numVecSample))
self.vecsRight[n] = np.ones((self.mps.virtual_dim[n+1], self.numVecSample))
