Question

这个神经网络的界面是作为不和谐机器人构建的。您可以在https://discord.gg/N5Bke3z找到它。如果在learn命令之前完成，reset命令可以正常工作。学习命令没有在我的消息监听器中响应。检查覆盖范围后，我注意到它正在跳过不可能的事情，例如if else语句。我使用的两个apis是javacord https://github.com/BtoBastian/Javacord，以及我处理数据的个人api https://github.com/NicksWorld/Networking-DataTypes。我的代码将丢失discord bot令牌，但我向你保证令牌不是问题。除了来自javacord的启动日志外，控制台日志为空：

2017年10月29日上午10:33:48 de.btobastian.javacord.utils.JavacordLogger info INFO：未找到与SLF4J兼容的记录器。使用默认值 javacord实现！

我很感激我能得到的任何帮助。我的代码位于https://github.com/NicksWorld/Java-neural-network，如下所示： Bot.java：

package Discord;

import com.google.common.util.concurrent.FutureCallback;

import Discord.message.Message;
import de.btobastian.javacord.DiscordAPI;
import de.btobastian.javacord.Javacord;

public class Bot {
    public static void main(String[] args) {
        //get login info
        DiscordAPI api = Javacord.getApi("*****************", true);
        //login
        api.connect(new FutureCallback<DiscordAPI>() {
            @Override
            public void onSuccess(final DiscordAPI api) {
                //set game and start listener
                api.setGame("Learning through Network001's algorithms");
                api.registerListener(new Message());
            }

            @Override
            public void onFailure(Throwable t) {
                t.printStackTrace();
            }
        });
    }
}

Message.java：

package Discord.message;

import de.btobastian.javacord.DiscordAPI;
import de.btobastian.javacord.listener.message.MessageCreateListener;
import me.NicksWorld.obj.DataCollection;
import me.NicksWorld.obj.IntegerRow;

public class Message implements MessageCreateListener{
    Network network = new Network();

    public IntegerRow StringToRow(String in) {
        String arr = in;
        String[] items = arr.replaceAll("\\[", "").replaceAll("\\]", "").replaceAll("\\s", "").split(",");
        IntegerRow results = new IntegerRow(items.length);
        for (int i = 0; i < items.length; i++) {
            try {
                results.set(i, Integer.parseInt(items[i]));
            } catch (NumberFormatException nfe) {
                nfe.printStackTrace();
            }
        }
        return results;
    }

    @Override
    public void onMessageCreate(DiscordAPI api, de.btobastian.javacord.entities.message.Message message) {
        //Stop interaction from bots
        if(message.getAuthor().isBot()) {
            return;
        }

        //register help command
        if (message.getContent().startsWith("!help")) {
            message.reply(message.getAuthor().getMentionTag() + "\n!help - Shows this list\n!learn - learns from a dataset in the form of an array ex. [1,2,3,4,5], that array tells it that it has the numbers 1, 2, 3, and 4. It also tells it that the result should be 5\n!find - takes an input of 4 numbers in an array ex. [1,2,3,4] so it can find an output based on conjectures from the training data");
        } else if(message.getContent().startsWith("!learn")) {
            if (network.learn(StringToRow(message.getContent().substring(7)))) {
                message.reply("Succes!");
            } else {
                message.reply("Fail :C");
            }
        } else if(message.getContent().startsWith("!find")) {

        } else if(message.getContent().startsWith("!reset")) {
            network.TrainingData = new DataCollection();
            message.reply("done");
        }
    }
}

Network.java：

package Discord.message;

import me.NicksWorld.obj.DataCollection;
import me.NicksWorld.obj.IntegerRow;

public class Network {
    //Initialize collection of training data
    public  DataCollection TrainingData = new DataCollection();
    //End initialize collection of training data

    //Initialize result variable
    public Double datasetResult = 0.0;
    //End initialize result variable

    //Initialize fails variable
    public Integer fails = 0;
    //End initialize fails variable

    //Initialize done learning boolean
    public boolean doneLearning = false;
    //End initialize done learning boolean

    //Initialize weights
    //Initialize column 1's weight
    public double ColumnWeight1 = Math.round(Math.random());
    //Initialize column 2's weight
    public double ColumnWeight2 = Math.round(Math.random());
    //Initialize column 3's weight
    public double ColumnWeight3 = Math.round(Math.random());
    //Initialize column 4's weight
    public double ColumnWeight4 = Math.round(Math.random());
    //End initialize weights


    //Function to check weights against all datasets
    public boolean checkWeights() {
        for (Integer indexOfTrainingData = 1; indexOfTrainingData <= TrainingData.get("integer").size(); indexOfTrainingData++) {
            //Reset variables for data
            datasetResult = 0.0;
            IntegerRow rowVar = (IntegerRow)TrainingData.get("integer").get(indexOfTrainingData - 1);
            //End reseting of variables

            //loop through the row
            for (Integer indexOfRow=1; indexOfRow <= 4; indexOfRow++) {
                //Determine which weight to use per value
                if (indexOfRow==1) {
                    datasetResult += ColumnWeight1 * rowVar.get().get(0);
                } else if (indexOfRow == 2) {
                    datasetResult += ColumnWeight2 * rowVar.get().get(1);
                } else if (indexOfRow == 3) {
                    datasetResult += ColumnWeight3 * rowVar.get().get(2);
                } else if (indexOfRow == 4) {
                    datasetResult += ColumnWeight4 * rowVar.get().get(3);
                }
            }
            if (datasetResult == rowVar.get().get(4).intValue()) {

            } else {
                return false;
            }
        }
        return true;
    }

    //Function to learn
    public Boolean learn(IntegerRow ToLearn) {
        //if(ToLearn.get().size()!=4) return false;
        //Add to training data list
        TrainingData.add(ToLearn);

        //loop through the training data
        fails = 0;
        for (Integer indexOfTrainingData = 1; indexOfTrainingData <= TrainingData.get("int").size(); indexOfTrainingData++) {
            //Reset variables for data
            datasetResult = 0.0;
            IntegerRow rowVar = (IntegerRow)TrainingData.get("int").get(indexOfTrainingData - 1);
            doneLearning = false;
            //End reseting of variables

            //determine when the for loop is complete
            while (!doneLearning) {
                //loop through the row
                for (Integer indexOfRow = 1; indexOfRow <= 4; indexOfRow++) {
                    //Determine which weight to use per value
                    if (indexOfRow==1) {
                        datasetResult += ColumnWeight1 * rowVar.get().get(0);
                    } else if (indexOfRow == 2) {
                        datasetResult += ColumnWeight2 * rowVar.get().get(1);
                    } else if (indexOfRow == 3) {
                        datasetResult += ColumnWeight3 * rowVar.get().get(2);
                    } else if (indexOfRow == 4) {
                        datasetResult += ColumnWeight4 * rowVar.get().get(3);
                    }
                }
                if (datasetResult == rowVar.get().get(4).intValue()) {
                    //check if successful with other datasets
                    if(checkWeights()) {
                        return true;
                    }
                } else {
                    fails++;
                    //Re-randomize weights
                    ColumnWeight1 = Math.round(Math.random());
                    ColumnWeight2 = Math.round(Math.random());
                    ColumnWeight3 = Math.round(Math.random());
                    ColumnWeight4 = Math.round(Math.random());
                }
            }
            return false;
        }
        return false;
    }
}

编辑：我通过查找失败计数未在正确的位置重置来修复错误。

Answer 1

我做了很多调试，并将问题精确定位到一行。失败计数正在重置，因为它停止了结束学习过程的失败计数。 Network.java的新代码是：

    package Discord.message;

import de.btobastian.javacord.DiscordAPI;
import de.btobastian.javacord.entities.message.Message;
import me.NicksWorld.obj.DataCollection;
import me.NicksWorld.obj.IntegerRow;

public class Network {
    //Initialize collection of training data

    public  DataCollection TrainingData = new DataCollection();

    //End initialize collection of training data


    //Initialize result variable

    public Double datasetResult = 0.0;

    //End initialize result variable


    //Initialize fails variable

    public Integer fails = 0;

    //End initialize fails variable


    //Initialize done learning boolean

    public boolean doneLearning = false;

    //End initialize done learning boolean


    //Initialize weights

    //Initialize column 1's weight
    public double ColumnWeight1 = Math.round(Math.random());
    //Initialize column 2's weight
    public double ColumnWeight2 = Math.round(Math.random());
    //Initialize column 3's weight
    public double ColumnWeight3 = Math.round(Math.random());
    //Initialize column 4's weight
    public double ColumnWeight4 = Math.round(Math.random());

    //End initialize weights


    //Function to check weights against all datasets

    public boolean checkWeights() {
        for(Integer indexOfTrainingData=1;indexOfTrainingData<=TrainingData.get("int").size();indexOfTrainingData++) {

            //Reset variables for data

            datasetResult = 0.0;
            IntegerRow rowVar = (IntegerRow) TrainingData.get("int").get(indexOfTrainingData - 1);

            //End reseting of variables


            //loop through the row

            for(Integer indexOfRow=1;indexOfRow<=4;indexOfRow++) {
                //Determine which weight to use per value
                if(indexOfRow==1) {
                    datasetResult += ColumnWeight1*rowVar.get().get(0);
                }else if(indexOfRow==2) {
                    datasetResult += ColumnWeight2*rowVar.get().get(1);
                }else if(indexOfRow==3) {
                    datasetResult += ColumnWeight3*rowVar.get().get(2);
                }else if(indexOfRow==4) {
                    datasetResult += ColumnWeight4*rowVar.get().get(3);
                }
            }
            if(datasetResult==rowVar.get().get(4).intValue()) {

            }else {
                return false;
            }
        }
        return true;
    }


    //Function to learn

    public Boolean learn(IntegerRow ToLearn, Message message) {
        if(ToLearn.get().size()!=5) {
            return false;
        }

        //Add to training data list

        TrainingData.add(ToLearn);


        //loop through the training data
        for(Integer indexOfTrainingData=1;indexOfTrainingData<=TrainingData.get("int").size();indexOfTrainingData++) {
            //Reset variables for data

            datasetResult = 0.0;
            IntegerRow rowVar = (IntegerRow) TrainingData.get("int").get(indexOfTrainingData - 1);
            doneLearning = false;
            //End reseting of variables

            //determine when the for loop is complete

            while(doneLearning != true) {
                //loop through the row
                datasetResult = 0.0;
                for(Integer indexOfRow=1;indexOfRow<=4;indexOfRow++) {
                    //Determine which weight to use per value
                    if(indexOfRow==1) {
                        datasetResult += ColumnWeight1*rowVar.get().get(0);
                    }else if(indexOfRow==2) {
                        datasetResult += ColumnWeight2*rowVar.get().get(1);
                    }else if(indexOfRow==3) {
                        datasetResult += ColumnWeight3*rowVar.get().get(2);
                    }else if(indexOfRow==4) {
                        datasetResult += ColumnWeight4*rowVar.get().get(3);
                    }
                }
                if(datasetResult==rowVar.get().get(4).intValue()) {
                    //check if successful with other datasets
                    Boolean test = checkWeights();
                    if(test==true) {
                        return true;
                    }
                }else {
                    fails++;
                    if(fails>1000) {
                        TrainingData = new DataCollection(); return false;
                    }

                    if(fails == 50|fails == 100|fails == 150|fails == 200|fails == 250) message.reply("Working...");
                    //Re-randomize weights
                    ColumnWeight1 = Math.round(Math.random());
                    ColumnWeight2 = Math.round(Math.random());
                    ColumnWeight3 = Math.round(Math.random());
                    ColumnWeight4 = Math.round(Math.random());
                }
            }
            TrainingData = new DataCollection();
            return false;
        }
        TrainingData = new DataCollection();
        return false;
    }
    //
    public Double findResult = 0.0;
    public Double find(IntegerRow in) {
        findResult = 0.0;

        for(Integer indexOfRow=1;indexOfRow<=4;indexOfRow++) {
            //Determine which weight to use per value
            if(indexOfRow==1) {
                findResult += ColumnWeight1*in.get().get(0);
            }else if(indexOfRow==2) {
                findResult += ColumnWeight2*in.get().get(1);
            }else if(indexOfRow==3) {
                findResult += ColumnWeight3*in.get().get(2);
            }else if(indexOfRow==4) {
                findResult += ColumnWeight4*in.get().get(3);
            }
        }
        return findResult;
    }
}

神经网络没有积极回应学习

1 个答案: