Using BPM Sparse

All replies

• 

  

  0

  Sign in to vote

  The following should give you some idea. Create a new console project, reference Infer.NET and the BayesPointMachine dll from building the BPM project. I have just shown code for the trainer. Do a similar thing for the test data except test data is not indexed by class.

  John

  using System;
  using System.Collections.Generic;
  using System.Linq;
  using System.Text;
  
  using MicrosoftResearch.Infer;
  using MicrosoftResearch.Infer.Distributions;
  using MicrosoftResearch.Infer.Factors;
  using MicrosoftResearch.Infer.Models;
  using BayesPointMachine;
  
  namespace willbuy
  {
  	class Program
  	{
  		static void Main(string[] args)
  		{
  			double[] incomes = { 63, 16, 28, 55, 22, 20 };
  			double[] ages = { 38, 23, 40, 27, 18, 40 };
  			bool[] willBuy = { true, false, true, true, false, false };
  
  			// Class 0 will be true, Class 1 will be false
  			int numData = incomes.Length;
  			int nClass = 2;
  			var indexList = new List<List<int>>[nClass];
  			var valueList = new List<List<double>>[nClass];
  			for (int classX = 0; classX < nClass; classX++)
  			{
  				indexList[classX] = new List<List<int>>();
  				valueList[classX] = new List<List<double>>();
  			}
  
  			// Build the data in form of indices and values
  			for (int dataX = 0; dataX < numData; dataX++)
  			{
  				List<int> indx = new List<int>();
  				List<double> vals = new List<double>();
  
  				// income index and value
  				indx.Add(0);
  				vals.Add(incomes[dataX]);
  
  				// age index and value
  				indx.Add(1);
  				vals.Add(ages[dataX]);
  
  				// bias input and value
  				indx.Add(2);
  				vals.Add(1.0);
  
  				// The label
  				int classX = willBuy[dataX] ? 0 : 1;
  				indexList[classX].Add(indx);
  				valueList[classX].Add(vals);
  			}
  
  			// Get data into array form
  			int[][][] indices = indexList.Select(c => c.Select(d => d.Select(i => i).ToArray()).ToArray()).ToArray();
  			double[][][] values = valueList.Select(c => c.Select(d => d.Select(v => v).ToArray()).ToArray()).ToArray();
  
  			BPMSparse bpmSparse = new BPMSparse(nClass, 3, 0.1);
  			var wInferred = bpmSparse.Train(indices, values);
  		}
  	}
  }

  
  

  Friday, March 9, 2012 9:15 PM

  Owner
• 

  

  0

  Sign in to vote

  Hi John,

  Thanks for the quick reply.  I started to look at your trainer above and the test version from the TestBPM project.  And added the below test set from the above code:

   double[] incomes = { 63, 16, 28, 55, 22, 20 };
              double[] ages = { 38, 23, 40, 27, 18, 40 };
              bool[] willBuy = { true, false, true, true, false, false };
  
              // Class 0 will be true, Class 1 will be false
              int numData = incomes.Length;
              int nClass = 2;
              var indexList = new List<List<int>>[nClass];
              var valueList = new List<List<double>>[nClass];
              for (int classX = 0; classX < nClass; classX++)
              {
                  indexList[classX] = new List<List<int>>();
                  valueList[classX] = new List<List<double>>();
              }
  
              // Build the data in form of indices and values
              for (int dataX = 0; dataX < numData; dataX++)
              {
                  List<int> indx = new List<int>();
                  List<double> vals = new List<double>();
  
                  // income index and value
                  indx.Add(0);
                  vals.Add(incomes[dataX]);
  
                  // age index and value
                  indx.Add(1);
                  vals.Add(ages[dataX]);
  
                  // bias input and value
                  indx.Add(2);
                  vals.Add(1.0);
  
                  // The label
                  int classX = willBuy[dataX] ? 0 : 1;
                  indexList[classX].Add(indx);
                  valueList[classX].Add(vals);
              }
  
              // Get data into array form
              int[][][] indices = indexList.Select(c => c.Select(d => d.Select(i => i).ToArray()).ToArray()).ToArray();
              double[][][] values = valueList.Select(c => c.Select(d => d.Select(v => v).ToArray()).ToArray()).ToArray();
  
              BPMSparse bpmSparse = new BPMSparse(nClass, 3, 0.1);
              var wInferred = bpmSparse.Train(indices, values);
  
             //test.............................................................
  
              int[][] indicesTestData = new int[2][];
  			double[][] valuesTestData = new double[2][];
  			indicesTestData[0] = new int[] { 0 };
  			indicesTestData[1] = new int[] { 1 };
              valuesTestData[0] = new double[] { 58, 18, 22 };
              valuesTestData[1] = new double[] { 36, 24, 37 };
  
        
  
              Console.WriteLine("\nPredictions:");
              Discrete[] predictions = bpmSparse.Test(indicesTestData, valuesTestData);
              foreach (Discrete pred in predictions)
                  Console.WriteLine(pred);
              
              Console.ReadLine();

  
  

  The output is:

  Discrete(0.9943 0.00566)
  Discrete(0.04148 0.9585)

  How do the above probabilities correspond to person 1, person 2, and person 3 from my test set?

  Monday, March 12, 2012 7:44 PM