package org.ea.javacnn.layers;

import org.ea.javacnn.data.BackPropResult;
import org.ea.javacnn.data.DataBlock;
import org.ea.javacnn.data.OutputDefinition;
import java.io.Serializable;
import java.util.ArrayList;
import java.util.List;

/**
 * This layer will remove some random activations in order to
 * defeat over-fitting.
 *
 * @author Daniel Persson (mailto.woden@gmail.com)
 */
public class DropoutLayer implements Layer,Serializable {
    private int out_depth, out_sx, out_sy;

    private DataBlock in_act, out_act;

    private final double drop_prob = 0.5;
    private boolean[] dropped;

    public DropoutLayer(OutputDefinition def) {
        // computed
        this.out_sx = def.getOutX();
        this.out_sy = def.getOutY();
        this.out_depth = def.getDepth();

        this.dropped = new boolean[this.out_sx*this.out_sy*this.out_depth];
    }

    @Override
    public DataBlock forward(DataBlock db, boolean training) {
        this.in_act = db;
        DataBlock V2 = db.clone();
        int N = db.getWeights().length;
        if(training) {
            // do dropout
            for(int i=0;i getBackPropagationResult() {
        return new ArrayList<>();
    }
}