I am trying to use numpy to perform a polyfit on a set of very large integers (~256bits). The points for the line are generated by randomly assigning coefficients and then picking X values at random. I then try to use numpy to recreate the polynomial from just those points, but the answer is not exact. It is usually off by less than 10, but it shouldn't be. Is there a way to turn up the precision for numpy? matlab is able to crunch these numbers and come up with the exact polynomial, so I would expect numpy to be able to do the same.