1: \begin{abstract}

2:   We propose in this work an original estimator of the conditional

3:   intensity of a marker-dependent counting process, that is, a

4:   counting process with covariates.  We use model selection methods

5:   and provide a non asymptotic bound for the risk of our estimator on

6:   a compact set. We show that our estimator reaches automatically a

7:   convergence rate over a functional class with a given (unknown)

8:   anisotropic regularity. Then, we prove a lower bound which

9:   establishes that this rate is optimal. Lastly, we provide a short

10:   illustration of the way the estimator works in the context of

11:   conditional hazard estimation.

12: \end{abstract}

13: