This paper presents joint transmit diversity selection (TDS) and relay selection (RS) algorithms based on discrete stochastic optimization for a multi-relay cooperative MIMO system. A two-phase, decode-and-forward (DF) network with a non-negligible direct path is considered where linear minimum mean square error (MMSE) receivers are used at all nodes. TDS and RS are performed jointly with continuous least squares channel estimation and no transmit preprocessing is required. RS removes relays from consideration by TDS and generates an optimized set from which TDS is made, improving the convergence, performance, complexity and energy consumption of the TDS process whilst maintaining low feedback requirements. The performance of the proposed schemes is evaluated via mean square error (MSE), bit-error rate (BER) and complexity comparisons. The results show that the proposed schemes outperform cooperative transmission with and without TDS in terms of diversity, MSE and BER, and match that of the optimum exhaustive solutions whilst making considerable complexity and energy savings.