Implicit conversion from float -5.675675675675677 to int loses precision