1: \begin{abstract} 

2: In some applications (e.g., in cosmology and economics), the regression $\E[Z|\vec{x}]$ is not adequate to represent the association between a predictor $\vec{x}$ and a response 

3:   $Z$  because of  multi-modality and asymmetry of $f(z|\vec{x})$; using the full density instead of a single-point estimate can then lead to less bias in subsequent analysis.  As of now, there are no effective ways of estimating $f(z|\vec{x})$ when $\vec{x}$ represents  high-dimensional, complex data. 

4: In this paper, we propose a new nonparametric estimator of $f(z|\vec{x})$ that adapts to sparse (low-dimensional) structure 

5: in $\vec{x}$.   By directly expanding $f(z|\vec{x})$ in the eigenfunctions of a kernel-based operator,  we avoid tensor products in high dimensions as well as ratios of estimated densities.

6:  Our basis functions are orthogonal with respect to the underlying data distribution, allowing fast implementation

7: and tuning of parameters. 

8: We derive rates of convergence and show that the method adapts to the intrinsic dimension of the data. We also demonstrate the effectiveness of the series method on images, spectra, and 

9:   an application to 

10:  photometric redshift estimation of galaxies.

11: 

12: 

13: 

14: 

15: 

16: %\keywords{spectral methods, nonparametric prediction and forecasting, series estimators, photometric redshift estimation} %, \red{others?}}

17: \end{abstract}

18: