Implicit conversion from float -21.476725521669337 to int loses precision