Decision tree using Accord.Net framework

Question

I'm trying to learn how to implement decision trees in C #. The examples I found are not helping me. What I want to do is offer resolutions for the problem that the user reports, I have a database with three thousand questions and answers. I am using dotnet and trying to do with the Accord.net framework, however I am not getting it, I need an example or tips on how to implement.

As I said, I have a database with the old questions and answers that will be used to train the algorithm and offer the possible answers to future problems.

P.S. I'm using visual studio community 2017.

Answer 1

The first thing that helps with understanding how to map this problem to a decision tree problem is this: each question should be treated as a split node (or a decision node) in the decision tree, and the collection of answers given to each question, helps with forming the child nodes of each split node. In this case of yours, we will treat each question as a column of a data table, and each row of the table as a set of answers provided to each question. The second thing to notice is that decision tree needs to predict "something", in your case it is the answer to one of the questions, we will consider that as the last column in the table. Since there was no written-example provided I assumed a medical data base, with a set of questions and a set of answers, as in the code below. You can easily map this to any other data type as long as it is also a set of questions and answers. The prediction column/question is the diagnosed disease in the example given below.

[ - This code is inspired by the examples provided @ http://accord-framework.net/docs/html/T_Accord_MachineLearning_DecisionTrees_Learning_ID3Learning.htm]

[- Prerequisite: you need to install Accord.Net packages: Install-Package Accord.MachineLearning]

using Accord.MachineLearning.DecisionTrees;
using Accord.MachineLearning.DecisionTrees.Learning;
using Accord.Math.Optimization.Losses;
using System;
using System.Data;
using Accord.Statistics.Filters;
using System.Linq;
using Accord.Math;

namespace DecisionTreeExamples
{
    public class QandADecisionTree{

    public static void QandADecisionTreeExample()
    {
        // Step1: Since there was no sample provided in this question, we first construct our data
        DataTable data = new DataTable("QandA");

        // Step2: Adding questions. Each question is a column in the data table. 
        PopulateQuestions(data); /*Note that the last column/question is the prediction column. */

        // Step3: Adding the history of 6 patients and the corresponding diagnosis (This is also our training data)
        PopulateAnswers(data);

        //Step4: Next we need to convert our data table to its numerical representation, which is what ID3 algorithm expects as input.
        Codification codification = new Codification(data);
        DataTable codifiedData = codification.Apply(data);

        //Note: specify all the columns except the last column, which is the value that the decision tree should predict, 
        int[][] input = codifiedData.ToJagged<int>("Are you above 50 years old?",
                                                   "What is your gender?",
                                                   "Do you have headache?",
                                                   "Has your blood pressure gone above 12 in the last 30 days?",
                                                   "Do you have a running nose?",
                                                   "Do you have a soat throat?",
                                                   "Do you have fever?");

        int[] predictions = codifiedData.ToArray<int>("Diagnosed Disease");

        // Step4: produce Decision tree.
        ID3Learning decisionTreeLearningAlgorithm = new ID3Learning { };

        DecisionTree decisionTree = decisionTreeLearningAlgorithm.Learn(input, predictions);

        // Step5: Use the produced decision tree on a test data
        int[] query = codification.Transform(new[,]{ {"Are you above 50 years old?","no" },
                                                   { "What is your gender?","male" },
                                                   { "Do you have headache?" ,"yes"},
                                                   { "Has your blood pressure gone above 12 in the last 30 days?","no" },
                                                   { "Do you have a running nose?","yes" },
                                                   { "Do you have a soat throat?" ,"yes"},
                                                   { "Do you have fever?" ,"yes"} });

        int result = decisionTree.Decide(query);
        string diagnosis = codification.Revert("Diagnosed Disease", result);

        Console.WriteLine($"Diagnosed disease: {diagnosis}"); // Prints Common Cold
    }

    /// <summary>
    /// Populates data table with questions, that in this example each question happens to be a column. 
    /// </summary>
    /// <param name="data">Specifies the data base in this example which is a collection of questions and answers.</param>
    private static void PopulateQuestions(DataTable data)
    {
        data.Columns.Add(new DataColumn("Are you above 50 years old?", typeof(string))); //Q1
        data.Columns.Add(new DataColumn("What is your gender?", typeof(string))); //Q2
        data.Columns.Add(new DataColumn("Do you have headache?", typeof(string))); //Q3
        data.Columns.Add(new DataColumn("Has your blood pressure gone above 12 in the last 30 days?", typeof(string))); //Q4
        data.Columns.Add(new DataColumn("Do you have a running nose?", typeof(string))); //Q5
        data.Columns.Add(new DataColumn("Do you have a soat throat?", typeof(string))); //Q6
        data.Columns.Add(new DataColumn("Do you have fever?", typeof(string))); //Q7
        data.Columns.Add(new DataColumn("Diagnosed Disease", typeof(string))); // Prediction Column
    }

    /// <summary>
    /// Populates data table with rows, that in this example each row happens to represent a patient, and each column, except the last one is a question. The value of a column in a row means the reponse patient provided to that question.
    /// </summary>
    /// <param name="data">Specifies the data base in this example which is a collection of questions and answers.</param>
    private static void PopulateAnswers(DataTable data)
    {
        var patient1 = data.NewRow();
        patient1["Are you above 50 years old?"] = "no";
        patient1["What is your gender?"] = "male";
        patient1["Do you have headache?"] = "yes";
        patient1["Has your blood pressure gone above 12 in the last 30 days?"] = "no";
        patient1["Do you have a running nose?"] = "yes";
        patient1["Do you have a soat throat?"] = "yes";
        patient1["Do you have fever?"] = "yes";
        patient1["Diagnosed Disease"] = "Common Cold";
        data.Rows.Add(patient1);

        var patient2 = data.NewRow();
        patient2["Are you above 50 years old?"] = "yes";
        patient2["What is your gender?"] = "female";
        patient2["Do you have headache?"] = "yes";
        patient2["Has your blood pressure gone above 12 in the last 30 days?"] = "yes";
        patient2["Do you have a running nose?"] = "no";
        patient2["Do you have a soat throat?"] = "no";
        patient2["Do you have fever?"] = "no";
        patient2["Diagnosed Disease"] = "High Blood Pressure";
        data.Rows.Add(patient2);


        var patient3 = data.NewRow();
        patient3["Are you above 50 years old?"] = "yes";
        patient3["What is your gender?"] = "male";
        patient3["Do you have headache?"] = "yes";
        patient3["Has your blood pressure gone above 12 in the last 30 days?"] = "yes";
        patient3["Do you have a running nose?"] = "no";
        patient3["Do you have a soat throat?"] = "no";
        patient3["Do you have fever?"] = "no";
        patient3["Diagnosed Disease"] = "High Blood Pressure";
        data.Rows.Add(patient3);

        var patient4 = data.NewRow();
        patient4["Are you above 50 years old?"] = "no";
        patient4["What is your gender?"] = "female";
        patient4["Do you have headache?"] = "yes";
        patient4["Has your blood pressure gone above 12 in the last 30 days?"] = "no";
        patient4["Do you have a running nose?"] = "yes";
        patient4["Do you have a soat throat?"] = "yes";
        patient4["Do you have fever?"] = "yes";
        patient4["Diagnosed Disease"] = "Common Cold";
        data.Rows.Add(patient4);

        var patient5 = data.NewRow();
        patient5["Are you above 50 years old?"] = "yes";
        patient5["What is your gender?"] = "female";
        patient5["Do you have headache?"] = "yes";
        patient5["Has your blood pressure gone above 12 in the last 30 days?"] = "no";
        patient5["Do you have a running nose?"] = "yes";
        patient5["Do you have a soat throat?"] = "yes";
        patient5["Do you have fever?"] = "yes";
        patient5["Diagnosed Disease"] = "Common Cold";
        data.Rows.Add(patient5);

        var patient6 = data.NewRow();
        patient6["Are you above 50 years old?"] = "yes";
        patient6["What is your gender?"] = "female";
        patient6["Do you have headache?"] = "no";
        patient6["Has your blood pressure gone above 12 in the last 30 days?"] = "yes";
        patient6["Do you have a running nose?"] = "no";
        patient6["Do you have a soat throat?"] = "no";
        patient6["Do you have fever?"] = "no";
        patient6["Diagnosed Disease"] = "High Blood Pressure";
        data.Rows.Add(patient6);
    }

}
}