Vector autoregressive (VAR) processes are useful in describing dynamical phenomena as diverse as speech, financial time-series, and the dancing of honey bees. However, such phenomena often exhibit structural changes over time and the VAR which describe them must also change. For example, the vocal tract of a speaker contracts; a country experiences a recession, a central bank intervention, or some national or global event; a honey bee changes from a waggle to a turn right dance. Some of these changes will appear fre- quently, while others are only rarely observed. In ad- dition, there is always the possibility of a previously unseen dynamic behavior. Thus, we propose a non- parametric approach for learning switching VAR pro- cesses, where we take the state sequence to be Markov....