After setting e.g. xlim to [.1, .9], the axes parameter editing window shows the xlims as 0.1000<...>1 and 0.9000<...>2 respectively, due to "classic" floating point repr issues. While this shoudl technically be fixed by numpy (numpy/numpy#6136) I wouldn't hold my breath for it and I'd suggest casting these items to Python floats, which use the shortest repr that yields the same value when eval'd.