1: \begin{abstract}

2: We establish minimax convergence rates for classification of functional

3: data and for nonparametric regression with functional design variables.

4: The optimal rates are of logarithmic type under smoothness constraints

5: on the functional density and the regression mapping, respectively.

6: These asymptotic properties are attainable by conventional kernel

7: procedures. The bandwidth selector does not require knowledge of the

8: smoothness level of the target mapping. In this work, the functional

9: data are considered as realisations of random variables which take

10: their values in a general Polish metric space. We impose certain metric

11: entropy constraints on this space; but no algebraic properties are required.

12: \end{abstract}