Implicit conversion from float -3.5064935064935128 to int loses precision