1: \begin{abstract}

2: In this paper we consider stochastic weakly convex composite problems, however

3: without the existence of a stochastic subgradient oracle. 

4: We present a derivative free algorithm that uses a two point

5: approximation for computing a gradient estimate of the smoothed 

6: function. We prove convergence at a similar rate

7: as state of the art methods, however with a larger constant, and report some numerical

8: results showing the effectiveness of the approach. 

9: \end{abstract}

10: