No implementation found for 'torch.index_select' on types that implement __torch_function__: [<class '__main__.Vertex'>]

1

I’m creating a new subclass using torch.Tensor and as mentioned in docs, overriding all the torch API using torch_function as below but still I’m getting above error. I checked torch.index_select is working if I run it directly on instance as torch.index_select(x, 0, torch.tensor([0,1])) considering x as a Vertex class.

class Vertex(torch.Tensor):
    def __init__(self, data):
        '''
        Theta shall be in radians. List is not supported yet.
        '''
        assert data.ndim in [1,2]
        assert len(data) == 4 if data.ndim==1 else len(data[0])==4, "Please follow g2o format"
        self.id = self[0]
        self.x = self[1]
        self.y = self[2]
        self.theta = self[3]
        super().__init__()
    
    @classmethod
    def __torch_function__(cls, func, types, args=..., kwargs=None):
        return super().__torch_function__(func, types, args, kwargs)

Any solution of this error?