Implicit conversion from float -20.099999999999994 to int loses precision