Implicit conversion from float -51.19999999999999 to int loses precision