Implicit conversion from float -0.6521739130434696 to int loses precision