Implicit conversion from float -36.13636363636364 to int loses precision