Implicit conversion from float -2.5490196078431495 to int loses precision