source: branches/3022-FastFunctionExtraction/TestFFX/Program.cs @ 18012

using System;
using System.IO;
using System.Collections.Generic;
using System.Globalization;
using System.Linq;
using HeuristicLab.Problems.DataAnalysis;
using System.Data;
using System.Diagnostics;
using System.Threading;
using HeuristicLab.Problems.DataAnalysis.Symbolic.Regression;

namespace HeuristicLab.Algorithms.DataAnalysis.FastFunctionExtraction {
    class Program {
        private static void TestPennML() {
            string projectDirectoryPath = Directory.GetParent(Environment.CurrentDirectory).Parent.FullName + "../branches/3022-FastFunctionExtraction/TestFFX/";
            string pennMLDataPath = projectDirectoryPath + "/pennML_data/";
            string targetPath = projectDirectoryPath + "/results/results.txt";
            var dInfo = new DirectoryInfo(pennMLDataPath);
            StreamWriter streamWriter = new StreamWriter(targetPath);
            streamWriter.WriteLine("dataset,algorithm,num_bases,train_mse,train_mae,test_mse,test_mae,runtime");
            streamWriter.Close();
            int len = dInfo.GetFiles().Length;
            int idx = 0;
            // foreach dataset
            foreach (var fInfo in dInfo.GetFiles()) {
                //if (fInfo.Name != "1089_USCrime.csv") continue;
                Console.Write($"{++idx}/{len}\t{fInfo.Name,-35}:");
                RunFFXOnDataset(fInfo, targetPath, out var elapsedSeconds);
                Console.WriteLine($"{elapsedSeconds + " sec",-15}");
            }
        }

        private static void SaveAccuracyInFile(ISymbolicRegressionSolution[] regressionSolutions, int[] numBases, double runtime, string filePath, string problemName) {
            if (filePath == "") return;
            CultureInfo culture = new CultureInfo("en-US");
            StreamWriter sw = new StreamWriter(filePath, true);
            int i = 0;
            foreach (var solution in regressionSolutions) {
                string outputStr = String.Join(",", new[]{
                    problemName.Substring(0, problemName.Length - 4),
                    "hl_ffx",
                    numBases[i++].ToString(),
                    solution.TrainingMeanSquaredError.ToString(culture),
                    solution.TrainingMeanAbsoluteError.ToString(culture),
                    solution.TestMeanSquaredError.ToString(culture),
                    solution.TestMeanAbsoluteError.ToString(culture),
                    runtime.ToString(culture)
                });
                sw.WriteLine(outputStr, culture);
            }
            sw.Dispose();
            Thread.Sleep(100); // to prevent race conditions with stream writers
        }

        private static void RunFFXOnDataset(FileInfo fInfo, string outFilePath, out double elapsedSeconds) {
            var data = ParseProblemDataFromFile(fInfo.FullName);
            Stopwatch sw = new Stopwatch();
            sw.Start();
            var regressionModels = FastFunctionExtraction.Fit(data, 0.95, out var numBases, true, true, true, true, true, maxNumBases: 10).ToArray();
            sw.Stop();
            var regressionSolutions = regressionModels.Select(model => new SymbolicRegressionSolution(model, data)).ToArray();
            elapsedSeconds = sw.Elapsed.TotalSeconds;
            SaveAccuracyInFile(regressionSolutions, numBases.ToArray(), elapsedSeconds, outFilePath, fInfo.Name);
        }

        private static IRegressionProblemData ParseProblemDataFromFile(string filepath, char separator = ';') {
            CultureInfo culture = new CultureInfo("en-US");
            var reader = new StreamReader(filepath);
            var variables = reader.ReadLine().Split(separator);
            var vals = Enumerable.Range(0, variables.Length).Select(_ => new List<double>()).ToArray();
            var targetVar = variables.Last();
            var allowedInputVars = variables.Where(val => val != targetVar);
            string line;

            line = reader.ReadLine();
            while (!String.IsNullOrWhiteSpace(line)) {
                int i = 0;
                foreach (var strVal in line.Split(separator)) {
                    vals[i++].Add(Convert.ToDouble(strVal, culture));
                }
                line = reader.ReadLine();
            }
            IDataset dataset = new Dataset(variables, vals);
            var temp = new RegressionProblemData(dataset, allowedInputVars, targetVar);
            return temp;
        }

        static void Main() {
            TestPennML();
            Console.WriteLine("Done.");
            Console.ReadLine();
        }
    }
}