MIMO transmit arrays allow for flexible design of the transmit beampattern. However, the large number of elements required to achieve certain performance using uniform linear arrays (ULA) maybe be too costly. This motivated the need for thinned arrays by appropriately selecting a small number of elements so that the full array beampattern is preserved. In this paper, we propose Learn-to-Select (L2S), a novel machine learning model for selecting antennas from a dense ULA employing a combination of multiple Softmax layers constrained by an orthogonalization criterion. The proposed approach can be efficiently scaled for larger problems as it avoids the combinatorial explosion of the selection problem. It also offers a flexible array design framework as the selection problem can be easily formulated for any metric.