Implicit conversion from float -64.2857142857143 to int loses precision