一种实用的降学习率公式：

             # set lr rate
            current_lr = 0.0002 * (1 / 2) ** (step / 10000)
            for param_group in optimizer_G.param_groups:
                param_group["lr"] = current_lr

具体使用方法：

     step = 0
    for epoch in range(opt.epoch, opt.n_epochs):
        for i, batch in enumerate(dataloader):
            step = step + 1

            # set lr rate
            current_lr = 0.0002 * (1 / 2) ** (step / 10000)
            for param_group in optimizer_G.param_groups:
                param_group["lr"] = current_lr