confuse about the calculate of least_square

[Rudin6]

`def least_square(from_p, to_p, dim=-1):
to_p = to_p.float()

beta = (to_p * from_p).sum(dim=dim, keepdim=True) / (from_p **2).sum(dim=dim, keepdim=True)

return beta

`
Why use the product form （to_p and from_p） instead of the subtraction form?

[LiyuanLucasLiu]

we are doing symmetric quantization, i.e., asking b to be 0 in wx + b. with some derivations in this setting, the least square has the above form : )