Implicit conversion from float -37.83783783783784 to int loses precision