Implicit conversion from float -3.9970717423133095 to int loses precision