This is the correct behavior, mpl tries to connect adjacent points. If you have points with a nan on either side, there is no valid way to connect them to anything.
If you want to just ignore your nans when plotting, then strip them out
ind = ~np.isnan(np.asarray(data.astype(float)))
plt.plot(np.asarray(x)[ind], np.asarray(data)[ind], 'ro--')
All the asarray are just to be sure the example will work, same with the astype
also, it is really bad practice to use variable names that shadow (both in terms of python resolution and in terms of semantic) classes (like array) or common functions.
