nntoolbox.init.lsuv module

Implement LSUV initialization from “ALL YOU NEED IS A GOOD INIT” https://arxiv.org/pdf/1511.06422.pdf Adopt from fastai

nntoolbox.init.lsuv.lsuv_init(module: torch.nn.modules.module.Module, input: torch.Tensor, tol: float = 0.001, Tmax: int = 100)

LSUV initialization :param module: :param input: :param tol: maximum tolerance :param Tmax: maximum iterations to attempt to demean and normalize weight :return: final mean and std of each layer’s output