Learning from multivariate discrete sequential data using a restricted Boltzmann machine model