Implicit conversion from float -3.4903047091412702 to int loses precision