Implicit conversion from float -21.188811188811187 to int loses precision