lib.py

from easydl import *


def reverse_sigmoid(y):
    return torch.log(y / (1.0 - y + 1e-10) + 1e-10)


def get_source_share_weight(domain_out, before_softmax, domain_temperature=1.0, class_temperature=10.0):
    before_softmax = before_softmax / class_temperature
    after_softmax = nn.Softmax(-1)(before_softmax)
    domain_logit = reverse_sigmoid(domain_out)
    domain_logit = domain_logit / domain_temperature
    domain_out = nn.Sigmoid()(domain_logit)
    
    entropy = torch.sum(- after_softmax * torch.log(after_softmax + 1e-10), dim=1, keepdim=True)
    entropy_norm = entropy / np.log(after_softmax.size(1))
    weight = entropy_norm - domain_out
    weight = weight.detach()
    return weight


def get_target_share_weight(domain_out, before_softmax, domain_temperature=1.0, class_temperature=10.0):
    return - get_source_share_weight(domain_out, before_softmax, domain_temperature, class_temperature)


def normalize_weight(x):
    min_val = x.min()
    max_val = x.max()
    x = (x - min_val) / (max_val - min_val)
    x = x / torch.mean(x)
    return x.detach()


def seed_everything(seed=1234):
    import random
    random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)
    np.random.seed(seed)
    import os
    os.environ['PYTHONHASHSEED'] = str(seed)