Implicit conversion from float -18.445945945945937 to int loses precision