how to train HMM with 8 states and 20 double numbers as input sequence?

[PARSA]

Hello everybody
I am working on a project to recognize human activities (8 activities) in real-time through body motion and HMM. I collect human data in the txt file as bellow with 20 double numbers:

(
0.2 0.3 0.2 0.6 1 2 5 0.8 1 1 0 0 0.4 0.3 0.2 0.2 0.6 1 2 5
....
....
...
...
.....
.....
)

I need to design a HMM that get as input a sequence with 20 double number and recognize one of the 8 activities as output in real time.

Does anyone have any sample code or at least a tutorial to learn how it can be done?
I have all data and I need to know how to train network with multiple states and continues data. The tutorial is not really well-explained to understand the process of training and using HMM.

Thanks
Regards

[César]

Hi Parsa,

Some months ago I published a tutorial for HMMs for mouse gesture recognition. It is just a "smaller" and simpler version of what you are trying to achieve. In your project, you are using 20 numbers. In the tutorial I wrote, I was using 2 (the x and y screen coordinates). But it should be not much complicated to expand it to incorporate more features. I have actually used the same mechanism to perform sign recognition in sign languages.

The tutorial is available at http://www.codeproject.com/Articles/541428/Sequence-Classifiers-in-Csharp-Part-I-Hidden-Marko

Hope it helps!

Best regards,

Cesar

[PARSA]

following your tutorial code I wrote my code but it gets error on this line  "double error = teacher.Run(inputs, outputs);" 

do you know how I can solve it?

private void Button_Click(object sender, RoutedEventArgs e)

 {          

            //Defining Input and output

  

          // hand movement from top to buttom

            double[] fvA1 = new double[] { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 };

            double[] fvA2 = new double[] { 2, 3, 4, 5, 6, 7, 8, 9, 10, 10 };

            double[] fvA3 = new double[] { 4, 3, 1, 2, 5, 6, 7, 8, 9, 10 };

            double[] fvA4 = new double[] { 6, 7, 8, 9, 10, 1, 2, 3, 4, 5 };

            double[][] action1 = { fvA1, fvA2, fvA3, fvA4 };

            // hand movement from left to right

            double[] fvB1 = new double[] { 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.10 };

            double[] fvB2 = new double[] { 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.10, 0.10 };

            double[] fvB3 = new double[] { 0.4, 0.3, 0.1, 0.2, 0.5, 0.6, 0.7, 0.8, 0.9, 0.10 };

            double[] fvB4 = new double[] { 0.6, 0.7, 0.8, 0.9, 0.10, 0.1, 0.2, 0.3, 0.4, 0.5 };

            double[][] action2 = { fvB1, fvB2, fvB3, fvB4 };

            // hand movement from top to buttom

            double[] fvC1 = new double[] { -1, -2, -3, -4, -5, -6, -7, -8, -9, -10 };

            double[] fvC2 = new double[] { -2, -3, -4, -5, -6, -7, -8, -9, -10, -10 };

            double[] fvC3 = new double[] { -4, -3, -1, -2, -5, -6, -7, -8, -9, -10 };

            double[] fvC4 = new double[] { -6, -7, -8, -9, -10, -1, -2, -3, -4, -5 };

            double[][] action3 = { fvC1, fvC2, fvC3, fvC4 };

            // hand movement from left to right

            double[] fvD1 = new double[] { 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.10 };

            double[] fvD2 = new double[] { 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.10, 0.10 };

            double[] fvD3 = new double[] { 0.4, 0.3, 0.1, 0.2, 0.5, 0.6, 0.7, 0.8, 0.9, 0.10 };

            double[] fvD4 = new double[] { 0.6, 0.7, 0.8, 0.9, 0.10, 0.1, 0.2, 0.3, 0.4, 0.5 };

            double[][] action4 = { fvD1, fvD2, fvD3, fvD4 };

            double[][][] inputs = new double[4][][] {action1, action2, action3, action4 };

            

            int[] outputs = new int[] 

            {

                0,

                1,

                2,

                3

            };

            

            

            string[] classes = new string[]{"A","B","C","D"};

            int states = 4;

            int iterations = 0;

            double tolerance = 0.01;

            bool rejection = false;

            hmm = new HiddenMarkovClassifier<MultivariateNormalDistribution>(4,

                new Forward(states), new MultivariateNormalDistribution(2), classes);

            // Create the learning algorithm for the ensemble classifier

            var teacher = new HiddenMarkovClassifierLearning<MultivariateNormalDistribution>(hmm,

                // Train each model using the selected convergence criteria

                i => new BaumWelchLearning<MultivariateNormalDistribution>(hmm.Models[i])

                {

                    Tolerance = tolerance,

                    Iterations = iterations,

                    FittingOptions = new NormalOptions()

                    {

                        Regularization = 1e-5

                    }

                }

            );

            teacher.Empirical = true;

            teacher.Rejection = rejection;

            // Run the learning algorithm

            double error = teacher.Run(inputs, outputs);

            double result = hmm.Compute(fvC1);

            Console.WriteLine("Result > " + result);

        } 

[César]

Hi Parsa,

From what I can see, it seems that you are using 10-dimensional feature vectors, but the code is specifying a multivariate Normal distributions with only 2 dimensions. I haven't tested yet, but if this is the cause then perhaps you would need to change

new MultivariateNormalDistribution(2)

to

new MultivariateNormalDistribution(10)

and see if it works better.

I hope it helps!

Best regards,

Cesar

[PARSA]

You are right, it resolved but another error on result. I am using last version of Accord 2,9

 double[] testinput = new double[] { -1, -2, -1, -1, 2, -6, -7, -0.8, -9, -10 };

            double result = hmm.Compute(testinput);

any idea about that?

[PARSA]

Thank you very much

The problem resolved. It was in input and output dimentions;

[César]

Hi Parsa,

Please accept my apologies for the late response; I couldn't answer it on time. I am glad you could solve the issue!

Best regards,

Cesar