浏览代码

Reorder operations

/goal-conditioning/new
vincentpierre 4 年前
当前提交
04bdb40c
共有 1 个文件被更改,包括 1 次插入1 次删除
  1. 2
      ml-agents/mlagents/trainers/torch/layers.py

2
ml-agents/mlagents/trainers/torch/layers.py


bound = math.sqrt(1 / (layer_size * self.input_size))
flat_output.weight.data.uniform_(-bound, bound)
self.hypernet = torch.nn.Sequential(*layers, flat_output, LayerNorm())
self.hypernet = torch.nn.Sequential(*layers, LayerNorm(), flat_output)
# The hypernetwork will not generate the bias of the main network layer
self.bias = torch.nn.Parameter(torch.zeros(output_size))

正在加载...
取消
保存