Implicit conversion from float -37.28571428571428 to int loses precision