• Create mines.mat from spreadsheet listing all cells.

  • Create exceptions.mat.

  • Run processgrid(0,0) in callanes.

  • load DataAccess\features.mat

  • Select the features you want to include in the network

    f = features([1:9 15]);

  • Select the set of target and don't-care classes for which you want to produce outputs from the net.

    targets = {'AT' 'AP'};
    dontcares = {'AP' 'AT'};

  • Make the network training dataset from the specified features.
    dataset = make_fowa_dataset(f,targets, dontcares);

    save('dataset','dataset');

  • Train the network.

    [Y, net] = train_fowa(dataset, 600);

  • Save your information.

    save('net-info','net','Y','f','targets','dontcares')

  • Check out the performance of your network:

    display_fowa_output(Y, dataset, net.owa_weights, 1, [0 0 .5 .5]);

    Argument 3 is 1 if we are to create omnibus output, and argument 4 is the vector of omnibus output weights (for weighting the min to max confidence output for front/back - left/right).

  • Run crossvalidation on the network:

    [crossval_error, crossval_separ, crossval_Y, avg_owas, nets, median_owas] = crossvalidate_fowa(dataset,num_models,num_epochs,output_weights,do_display,name,init_nets)

    Right now, we use output_weights like [ 0 0 .5 .5 ] or [ 0 0 0 1 ].

  • Copy the following files to the blindlane directory:

    • WhitenMatFromBlank1.mat
    • net-info.mat

  • Change directory to the blindlane and make the blind dataset

    dataset = make_fowa_dataset(f,targets, dontcares);
    save('dataset','dataset')

  • Run the network on the blinddataset:

    Y = computefowa(dataset, net);

  • Convert the net output into a features file:

    fowa_feature_struct(Y) ???????