def __init__(self, para=Parameters.gtn(), debug_mode=False, device='cpu'):
# Initialize Parameters
Programclass.Program.__init__(self, device=device, dtype=para['dtype'])
MLclass.MachineLearning.__init__(self, para, debug_mode)
MPSclass.MPS.__init__(self)
self.initialize_parameters_gtn()
self.name_md5_generate()
self.debug_mode = debug_mode
# Initialize MPS and update info
if not debug_mode:
self.load_gtn()
# if not exist saved mps gtn model,then initialize
if len(self.tensor_data) == 0:
self.initialize_dataset()
# Initialize info
self.generate_tensor_info()
self.generate_update_info()
self.initialize_mps_gtn()
if not self.tensor_data[0].device == self.device:
for ii in range(len(self.tensor_data)):
self.tensor_data[ii] = torch.tensor(self.tensor_data[ii],
device=self.device)
if device == 'cuda': torch.cuda.empty_cache()
# Environment Preparation
self.tensor_input = tuple()
self.environment_left = tuple()
self.environment_right = tuple()
self.environment_zoom = tuple()
from library import TNMLclass from library import Parameters as Pa pa = Pa.gtn() A = TNMLclass.GTN(para=pa, device='cuda:0', debug_mode=False) A.start_learning()
from library import MPSMLclass
from library import Parameters
para = Parameters.gtn()
GTN = MPSMLclass.GTN(para=para,
device='cpu') # change device='cuda' to use GPU
GTN.start_learning()