Implicit conversion from float -53.25 to int loses precision