mindchemistry.e3.o3.Linear

查看源文件
class mindchemistry.e3.o3.Linear(irreps_in, irreps_out, ncon_dtype=float32, **kwargs)[源代码]

线性等变操作。 等效于"instructions='linear'"的"TensorProduct'"。有关详细信息,请参阅 mindchemistry.e3.o3.TensorProduct

参数:
  • irreps_in (Union[str, Irrep, Irreps]) - 输入的Irreps。

  • irreps_out (Union[str, Irrep, Irreps]) - 输出的Irreps。

  • irrep_norm (str) - {'component','norm'},输入和输出表示的假定归一化。默认值: "component"

  • path_norm (str) - {'element','path'},路径权重的规范化方法。默认值:'element'

  • weight_init (str) - {'zeros','ones','truncatedNormal','normal','uniform','he_uniform','she_normal','xavier_uniform'},权重的初始方法。默认值:"normal"

  • ncon_dtype (mindspore.dtype) - ncon 计算模块输入张量的类型。默认值:mindspore.float32

支持平台:

Ascend

样例:

>>> from mindchemistry.e3.o3 import Linear
>>> Linear('2x2e+3x1o+3x0e', '3x2e+5x1o+2x0e')
TensorProduct [linear] (2x2e+3x1o+3x0e x 1x0e -> 3x2e+5x1o+2x0e)