Let's do the time-warp-attend: Learning topological invariants of dynamical systems