def __init__(self, args, test_dataloader):
self.args = args
self.test_dataloader = test_dataloader
self.bce_loss = nn.BCELoss()
self.mse_loss = nn.MSELoss()
self.ce_loss = nn.CrossEntropyLoss()
self.sampler = sampler.AdversarySampler(self.args.budget)
def __init__(self, args, test_dataloader):
self.args = args
self.test_dataloader = test_dataloader
self.model_path = self.args.out_path
self.bce_loss = nn.BCELoss()
self.mse_loss = nn.MSELoss()
self.ce_loss = nn.CrossEntropyLoss()
# self.base_name = "discriminator_probs_" # Only in the discriminator probabilties changes, no backward vae
# self.base_name = "vae_back_dis_probs_" # Vae backward done and discriminator probabilities only
self.base_name = "m2_model"
self.sampler = sampler.AdversarySampler(self.args.budget)
if args.tensorboard is True:
tb_path = os.path.join(args.out_path, 'tb_logs')
if not os.path.exists(tb_path):
os.mkdir(tb_path)
self.writer_train = SummaryWriter(os.path.join(tb_path, 'summary_train'))
self.writer_val = SummaryWriter(os.path.join(tb_path, 'summary_val'))
def __init__(self, args, test_dataloader):
self.args = args
self.test_dataloader = test_dataloader
self.bce_loss = nn.BCELoss()
self.mse_loss = nn.MSELoss()
self.ce_loss = nn.CrossEntropyLoss()
self.sampling_method = args.sampling_method
if self.sampling_method == "random":
self.sampler = sampler.RandomSampler(self.args.budget)
elif self.sampling_method == "adversary":
self.sampler = sampler.AdversarySampler(self.args.budget)
elif self.sampling_method == "uncertainty":
self.sampler = sampler.UncertaintySampler(self.args.budget)
elif self.sampling_method == "expected_error":
self.sampler = sampler.EESampler(self.args.budget)
elif self.sampling_method == "adversary_1c":
self.sample = sampler.AdversarySamplerSingleClass(self.args.budget)
else:
raise Exception("No valid sampling method provideds")