Implicit conversion from float -21.23076923076924 to int loses precision