Implicit conversion from float -20.849999999999994 to int loses precision