Implicit conversion from float -50.71428571428572 to int loses precision