Hmm = new HiddenMarkovClassifier<MultivariateMixture<MultivariateNormalDistribution>>(classes.Count,
                new Forward(states), new MultivariateMixture<MultivariateNormalDistribution>(new MultivariateNormalDistribution[6]), classes.ToArray());
            // I tried to change the topology from Forward to Ergodic and the performence is nearly the same
            // Ergodic models are commonly used to represent models in which a single (large) sequence of observations is used for training (such as when a training sequence does not have well defined starting and ending points and can potentially be infinitely long).
            // Create the learning algorithm for the ensemble classifier
            var teacher = new HiddenMarkovClassifierLearning<MultivariateMixture<MultivariateNormalDistribution>>(Hmm,

                // Train each model using the selected convergence criteria
                i => new BaumWelchLearning<MultivariateMixture<MultivariateNormalDistribution>>(Hmm.Models[i])
                {
                    Tolerance = tolerance,
                    Iterations = iterations,

                    FittingOptions = new NormalOptions()
                    {
                        Regularization = 1e-5
                       // , Diagonal = true

                    }
                }
            );