This page contains additional information regarding the HMM Annotator implemented in SPINE1.3
First, we provide a brief overview of the Markov processes and the Hidden Markov Model idea behind our implementation in HMM Overview. The document also contains a brief overview of the method used in the implementation.
Our implementation is based on collecting training data. The SPINE 1.3 release includes a training data set. Some users may find it not working for their purposes. In order to solve this issue we are providing the code for system training based on the input data. Documentation about the application can be found at HMM Training Work Flow.
The training is implemented in MATLAB. Since not everyone has access to MATLAB we converted our code to an executable that does not require a full MATLAB installation HMM install. the archive consists of two files
- setup_hmm_pkg - sets up the MATLAB runtime environment, sets up the folders required by the HMM training program, unpacks a set of training data into those directories. The system can be tested with the following parameters based on the test data.
- Experiment 1
- Subject 1
- Movement 1,2
- Nodes 3,4
- Yes, use pressure annotations
- Node 18 for pressure annotations
- 0.5 samples used for training
- Any file name you want to have the result in. For example test.txt.
- hmm_train - provides a GUI asking for the parameters discussed above as input. Provides .h training files for each one of the nodes used in training. For the example of "test.txt" and 2 input nodes, there will be two output files generated test3.txt and test4.txt