Spatially adaptive density estimation by localised Haar projections

Gach, Florian ; Nickl, Richard ; Spokoiny, Vladimir

Annales de l'I.H.P. Probabilités et statistiques, Tome 49 (2013), p. 900-914 / Harvested from Numdam

Résumé

A partir d’un échantillon d’une loi de densité $f_{0}:\mathbb {R}\rightarrow [0,\infty )$, nous construisons des estimateurs par ondelettes de Haar de $f_{0}$, dont les niveaux de résolution varient et sont construits à partir de tests localisés (comme dans l’article Lepski (Ann. Statist. 25 (1997) 927-947)). Nous montrons que ces estimateurs satisfont une inégalité oracle adaptive par rapport à la régularité potentiellement hétérogène de $f_{0}$, simultanément pour tout point $x$ dans un intervalle donné, en norme infinie. Les constantes de seuillage utilisées dans les procédures de test peuvent être choisies en pratique en supposant de manière idéalisée que la vraie densité est localement constante dans un voisinage du point $x$ considéré, pratique que nous justifions par un argument de théorie de l’information.

Abstract

Given a random sample from some unknown density $f_{0}:\mathbb {R}\rightarrow [0,\infty )$ we devise Haar wavelet estimators for $f_{0}$ with variable resolution levels constructed from localised test procedures (as in Lepski, Mammen and Spokoiny (Ann. Statist. 25 (1997) 927-947)). We show that these estimators satisfy an oracle inequality that adapts to heterogeneous smoothness of $f_{0}$, simultaneously for every point $x$ in a fixed interval, in sup-norm loss. The thresholding constants involved in the test procedures can be chosen in practice under the idealised assumption that the true density is locally constant in a neighborhood of the point $x$ of estimation, and an information theoretic justification of this practise is given.

Détail

Publié le : 2013-01-01

MR 3112439

DOI : https://doi.org/10.1214/12-AIHP485
Classification:  62G05

Bibliographie

[1] D. L. Donoho and I. M. Johnstone. Ideal spatial adaptation by wavelet shrinkage. Biometrika 81 (1994) 425-455. | MR 1311089 | Zbl 0815.62019

[2] D. L. Donoho, I. M. Johnstone, G. Kerkyacharian and D. Picard. Wavelet shrinkage: Asymptopia? J. R. Stat. Soc. Ser. B Stat. Methodol. 57 (1995) 301-369. | MR 1323344 | Zbl 0827.62035

[3] D. L. Donoho, I. M. Johnstone, G. Kerkyacharian and D. Picard. Density estimation by wavelet thresholding. Ann. Statist. 24 (1996) 508-539. | MR 1394974 | Zbl 0860.62032

[4] E. Giné, R. Latala and J. Zinn. Exponential and moment inequalities for $U$-statistics. In High Dimensional Probability II 13-38. E. Giné, D. Mason and J. A. Wellner (Eds). Birkhäuser Boston, Boston, MA, 2000. | MR 1857312 | Zbl 0969.60024

[5] E. Giné and R. Nickl. An exponential inequality for the distribution function of the kernel density estimator, with applications to adaptive estimation. Probab. Theory Related Fields 143 (2009) 569-596. | MR 2475673 | Zbl 1160.62032

[6] E. Giné and R. Nickl. Uniform limit theorems for wavelet density estimators. Ann. Probab. 37 (2009) 1605-1646. | MR 2546757 | Zbl 1255.62103

[7] E. Giné and R. Nickl. Adaptive estimation of a distribution function and its density in sup-norm loss by wavelet and spline projections. Bernoulli 16 (2010) 1137-1163. | MR 2759172 | Zbl 1207.62082

[8] A. Goldenshluger and O. Lepski. Structural adaptation via $\mathbb {L}_{p}$-norm oracle inequalities. Probab. Theory Related Fields 143 (2009) 41-71. | MR 2449122 | Zbl 1149.62020

[9] S. Jaffard. On the Frisch-Parisi conjecture. J. Math. Pures Appl. 79 (2000) 525-552. | MR 1770660 | Zbl 0963.28009

[10] O. V. Lepski, E. Mammen and V. Spokoiny. Optimal spatial adaptation to inhomogeneous smoothness: An approach based on kernel estimates with variable bandwidth selectors. Ann. Statist. 25 (1997) 929-947. | MR 1447734 | Zbl 0885.62044

[11] V. Spokoiny and C. Vial. Parameter tuning in pointwise adaptation using a propagation approach. Ann. Statist. 37 (2009) 2783-2807. | MR 2541447 | Zbl 1173.62028