AlanVey
commented
8 years ago @SX729 It depends which technique you wish to use:
- ANN: Under subfolder NN you will find 4 different techniques used to train the network. Running the function in optParams will generate optimal parameters by doing a very simple search. Once you have your parameters and have chosen a training method just use the corresponding create network function and use nFoldCrossValidation to train it. You can then use the functions in the root directory for ANN to predict based on the clean or noisy data sets and calculate various statistics.
- DT: Use nFoldCrossValidation to generate predictions and trees (one for each binary label). Note since we have 6 emotions it will generate six trees. Care needs to be taken as to how you deal with a data point being classified as multiple emotions or none.
- CBR: In theory this should have been just as good as the other two implementations however we were not able to replicate those results here so I won't go into detail on this.
Could you tell me how to run the code？Thank you very much！