.apply(np.linalg.norm, axis=1)
Is probably preferable since it is clearer, but when I'm just doing scratch work I often use:
df[list('xyz')].pow(2).sum(axis=1).pow(.5)