network.aika.lattice.InputNode Maven / Gradle / Ivy

Go to download
/*
 * Licensed to the Apache Software Foundation (ASF) under one or more
 * contributor license agreements.  See the NOTICE file distributed with
 * this work for additional information regarding copyright ownership.
 * The ASF licenses this file to You under the Apache License, Version 2.0
 * (the "License"); you may not use this file except in compliance with
 * the License.  You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
package network.aika.lattice;


import network.aika.Document;
import network.aika.Model;
import network.aika.Provider;
import network.aika.neuron.Neuron;
import network.aika.neuron.activation.Position;
import network.aika.neuron.relation.Relation;
import network.aika.neuron.activation.Activation;
import network.aika.lattice.AndNode.AndActivation;
import network.aika.lattice.InputNode.InputActivation;
import network.aika.lattice.AndNode.Refinement;
import network.aika.lattice.AndNode.RefValue;
import network.aika.lattice.AndNode.RelationsMap;

import java.io.DataInput;
import java.io.DataOutput;
import java.io.IOException;
import java.util.*;


/**
 * The {@code InputNode} class is the input layer for the boolean logic. The input-node has two sources of
 * activations. First, it might be underlying logic node of an {@code InputNeuron} in which case the input
 * activations come from the outside. The second option is that the activation come from the output of another neuron.
 *
 * @author Lukas Molzberger
 */
public class InputNode extends Node {

    public static int CHILD_NODE_THRESHOLD = 10;


    private Neuron inputNeuron;

    private TreeMap nonExactAndChildren;


    public InputNode() {
    }


    public InputNode(Model m) {
        super(m, 1);
    }


    public void setInputNeuron(Neuron n) {
        inputNeuron = n;
    }


    public void addActivation(Activation inputAct) {
//        if(inputAct.repropagateV != null && inputAct.repropagateV != markedCreated) return;

        InputActivation act = new InputActivation(inputAct, this);

        addActivation(act);
    }


    public void reprocessInputs(Document doc) {
        inputNeuron.get(doc).getActivations(doc, false).forEach(act -> {
//            act.repropagateV = markedCreated;
            if(act.upperBound > 0.0) {
                act.getINeuron().propagate(act);
            }
        });
    }


    void addAndChild(AndNode.Refinement ref, AndNode.RefValue child) {
        super.addAndChild(ref, child);

        if(!ref.relations.isExact()) {
            if (nonExactAndChildren == null) {
                nonExactAndChildren = new TreeMap<>();
            }

            AndNode.RefValue n = nonExactAndChildren.put(ref, child);
            assert n == null;
        }
    }


    void removeAndChild(AndNode.Refinement ref) {
        super.removeAndChild(ref);

        if(!ref.relations.isExact()) {
            if (nonExactAndChildren != null) {
                nonExactAndChildren.remove(ref);

                if (nonExactAndChildren.isEmpty()) {
                    nonExactAndChildren = null;
                }
            }
        }
    }


    public RefValue expand(int threadId, Document doc, Refinement ref) {
        if(!ref.isConvertible()) return null;

        Relation rel = ref.relations.get(0);
        if(rel == null) {
            return null;
        }

        RefValue rv = getAndChild(ref);
        if(rv != null) {
            return rv;
        }

        List nlParents = new ArrayList<>();

        Refinement mirrorRef = new Refinement(new AndNode.RelationsMap(new Relation[]{rel.invert()}), provider);
        nlParents.add(new AndNode.Entry(mirrorRef, new RefValue(new Integer[] {1}, 0, ref.input)));

        rv = new RefValue(new Integer[] {0}, 1, provider);
        nlParents.add(new AndNode.Entry(ref, rv));

        return AndNode.createAndNode(provider.model, doc, nlParents, level + 1) ? rv : null;
    }


    /**
     * @param act
     */
    @Override
    protected void propagate(InputActivation act) {
        try {
            lock.acquireReadLock();
            if (andChildren != null) {
                TreeMap children;
                if(andChildren.size() > CHILD_NODE_THRESHOLD) {
                    children = nonExactAndChildren;
                    propagateWithExactRelations(act);
                } else {
                    children = andChildren;
                }

                if(children != null) {
                    children.forEach((ref, rv) -> {
                        InputNode in = ref.input.getIfNotSuspended();
                        if (in != null) {
                            in.addNextLevelActivations(ref, rv.child.get(act.getDocument()), act);
                        }
                    });
                }
            }
        } finally {
            lock.releaseReadLock();
        }

        propagateToOrNode(act);
    }


    private void propagateWithExactRelations(InputActivation act) {
        Activation iAct = act.input;
        Document doc = act.getDocument();

        for (Map.Entry me : iAct.slots.entrySet()) {
            for (Activation linkedAct : act.getDocument().getActivationsByPosition(me.getValue(), true, me.getValue(), true)) {
                Provider in = linkedAct.getINeuron().getOutputNode();
                for (Map.Entry mea : andChildren.subMap(
                        new Refinement(RelationsMap.MIN, in),
                        new Refinement(RelationsMap.MAX, in)).entrySet()) {
                    in.get(doc).addNextLevelActivations(mea.getKey(), mea.getValue().child.get(doc), act);
                }
            }
        }
    }


    private void addNextLevelActivations(Refinement ref, AndNode nln, InputActivation act) {
        Document doc = act.getDocument();

        if(inputNeuron.get().isEmpty(doc)) return;

        Activation iAct = act.input;

        if(act.repropagateV != null && act.repropagateV != nln.markedCreated) return;

        ref.relations.get(0).getActivations(inputNeuron.get(doc), iAct)
                .filter(secondIAct -> secondIAct.getOutputNodeActivation() != null)
                .map(secondIAct -> secondIAct.getOutputNodeActivation())
                .filter(secondAct -> secondAct != null && secondAct.registered)
                .forEach(secondAct -> {
                    //    if (!Conflicts.isConflicting(iAct, secondIAct)) {
                    AndActivation oAct = new AndActivation(doc, nln);
                    for (AndNode.Entry e : nln.parents) {
                        boolean match = e.ref.compareTo(ref) == 0;
                        oAct.link(e.ref, e.rv, match ? secondAct : act, match ? act : secondAct);
                    }
                    nln.addActivation(oAct);
                    // }
                }
        );
    }


    @Override
    public void cleanup() {
    }


    public String logicToString() {
        StringBuilder sb = new StringBuilder();
        sb.append("I");

        sb.append("[");

        if (inputNeuron != null) {
            sb.append(inputNeuron.id);
            if (inputNeuron.getLabel() != null) {
                sb.append(",");
                sb.append(inputNeuron.getLabel());
            }
        }

        sb.append("]");

        return sb.toString();
    }



    @Override
    public void write(DataOutput out) throws IOException {
        out.writeBoolean(false);
        out.writeChar('I');
        super.write(out);

        out.writeBoolean(inputNeuron != null);
        if (inputNeuron != null) {
            out.writeInt(inputNeuron.id);
        }
    }


    @Override
    public void readFields(DataInput in, Model m) throws IOException {
        super.readFields(in, m);

        if (in.readBoolean()) {
            inputNeuron = m.lookupNeuron(in.readInt());
        }
    }


    public static class InputActivation extends NodeActivation {

        public Activation input;

        public InputActivation(Activation iAct, InputNode node) {
            super(iAct.getDocument(), node);
            input = iAct;
            iAct.setOutputNodeActivation(this);
        }

        public Activation getInputActivation(int i) {
            assert i == 0;
            return input;
        }


        public String toString() {
            return "I-ACT(" + input.getLabel() + " " + input.slotsToString() + ")";
        }
    }
}