mindspore网络构建调用numpy算子失败

我在进行Pytorch网络向mindspore迁移时，发现mindspore里缺少一个einsum算子。经过查阅资料发现，numpy中有类似的api。所以就想在网络构建时调用numpy中的

einsum算子替代。

我看mindspore文档中说是，在construct函数中不能调用第三方算子，于是我先在init函数中初始化了一个np.einsum对象，然后在construct函数中调用该算子。

class SpatialGraphConv(nn.Cell):
    def __init__(self, in_channels, out_channels, max_graph_distance):
        super(SpatialGraphConv, self).__init__()

        # spatial class number (distance = 0 for class 0, distance = 1 for class 1, ...)
        self.s_kernel_size = max_graph_distance + 1

        # weights of different spatial classes
        self.gcn = nn.Conv2d(in_channels=in_channels, out_channels=out_channels*self.s_kernel_size,
                             kernel_size=1, has_bias=True, pad_mode='valid')
        self.einsum = np.einsum

    def construct(self, x, A):
        # numbers in same class have same weight
        x = self.g