Implicit conversion from float -4.375 to int loses precision