This paper proposes a multitask learning method of joint optimization based on sparse linear array position of elements and weights of the array. The method can get more information from the target pattern features without adequate prior knowledge, and convert the sparse array antennas into a linear regression problem sparse matrix. By identifying sparse vector support area, the paper transforms the underdetermined steering vector matrix equation to the over determined conversion feature matrix equation. The results show that the proposed algorithm can optimize the position and weights of elements, and at the same time, this can realize the shaped beam pattern effectively.