Free cookie consent management tool by TermsFeed Policy Generator

Changeset 15924


Ignore:
Timestamp:
04/30/18 20:24:17 (6 years ago)
Author:
gkronber
Message:

#2886 remove obsolete code in C# program for the evaluation of sentences, switch to NSME as quality measure. Tried plotting functions within clusters in R

Location:
branches/2886_SymRegGrammarEnumeration
Files:
3 edited

Legend:

Unmodified
Added
Removed

  • branches/2886_SymRegGrammarEnumeration/ExpressionClustering/ExpressionClustering.csproj

    r15903 r15924  
    3232    <ErrorReport>prompt</ErrorReport>
    3333    <WarningLevel>4</WarningLevel>
     34  </PropertyGroup>
     35  <PropertyGroup Condition="'$(Configuration)|$(Platform)' == 'Debug|x64'">
     36    <DebugSymbols>true</DebugSymbols>
     37    <OutputPath>bin\x64\Debug\</OutputPath>
     38    <DefineConstants>DEBUG;TRACE</DefineConstants>
     39    <DebugType>full</DebugType>
     40    <PlatformTarget>x64</PlatformTarget>
     41    <ErrorReport>prompt</ErrorReport>
     42    <CodeAnalysisRuleSet>MinimumRecommendedRules.ruleset</CodeAnalysisRuleSet>
     43    <Prefer32Bit>true</Prefer32Bit>
     44  </PropertyGroup>
     45  <PropertyGroup Condition="'$(Configuration)|$(Platform)' == 'Release|x64'">
     46    <OutputPath>bin\x64\Release\</OutputPath>
     47    <DefineConstants>TRACE</DefineConstants>
     48    <Optimize>true</Optimize>
     49    <DebugType>pdbonly</DebugType>
     50    <PlatformTarget>x64</PlatformTarget>
     51    <ErrorReport>prompt</ErrorReport>
     52    <CodeAnalysisRuleSet>MinimumRecommendedRules.ruleset</CodeAnalysisRuleSet>
     53    <Prefer32Bit>true</Prefer32Bit>
    3454  </PropertyGroup>
    3555  <ItemGroup>

  • branches/2886_SymRegGrammarEnumeration/ExpressionClustering/Program.cs

    r15903 r15924  
    22using System.Collections;
    33using System.Collections.Generic;
    4 using System.Drawing;
     4using System.Diagnostics;                                                                                             
    55using System.IO;
    6 using System.Linq;
    7 using HeuristicLab.Analysis;
    8 using HeuristicLab.Analysis.Views;
    9 using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
     6using System.Linq;                                                                                                     
    107using HeuristicLab.Problems.DataAnalysis;
    118using HeuristicLab.Problems.DataAnalysis.Symbolic;
    129
    13 // Evaluates sentences on randomly generated data
     10// Reads sentences from files, determines the shortest infix expression for a hash and evaluates sentences on randomly generated data, evaluation result is written to file
    1411namespace ExpressionClustering {
    1512  class Program {
    16     private static readonly string folder = Environment.GetFolderPath(Environment.SpecialFolder.DesktopDirectory);
    17     private static readonly string clusterFolder = Path.Combine(Environment.GetFolderPath(Environment.SpecialFolder.DesktopDirectory), "clusters");
    18     private static readonly string distinctSentencesFileName = Path.Combine(folder, @"distinctSentences_2018-04-13_09-52_TreeSize-10.csv");
    19     private static readonly string allSentencesFileName = Path.Combine(folder, "allSentences_2018-04-13_09-52_TreeSize-10.csv");
    20     private static readonly string outputFileName = Path.Combine(folder, "evaluations_2018-04-13_09-52_TreeSize-10.csv.gz");
     13    //private static readonly string folder = @"D:\heal\documents\trunk\Publications\2018\GPTP\data";
     14    //private static readonly string clusterFolder = folder;
     15    //private static readonly string distinctSentencesFileName = Path.Combine(folder, @"distinctSentences_2018-04-13_16-40_TreeSize-7.csv.gz");
     16    //private static readonly string allSentencesFileName = Path.Combine(folder, "allSentences_2018-04-16_14-49_TreeSize-8_1d.csv.gz");
     17    //private static readonly string outputFileName = Path.Combine(folder, "evaluations_2018-04-16_14-49_TreeSize-8_1d.csv.gz");
    2118    private static int N = 100;
     19    private static int PERF_STATS_UPDATE_INTERVAL = 100000;
    2220    private static double[] evalBuf = new double[N];
    2321
    … …  
    3634      .ToArray();
    3735
    38 
    39     public static int MAX_STACK = 20;
    40     public static double[][] stack = new double[MAX_STACK][];
    41     static Program() {
    42       for (int i = 0; i < MAX_STACK; i++)
    43         stack[i] = new double[N];
    44     }
    45 
     36   
    4637    // loads symbolic expressions in postfix notation from a stream and identifies clusters of expressions
    4738    static void Main(string[] args) {
     39      var sentencesFileName = args[0];
     40      var outputFileName = Path.Combine(Path.GetDirectoryName(sentencesFileName), "evaluations_" + Path.GetFileName(sentencesFileName));
     41                               
    4842
    49       var hash2Postfix = new Dictionary<string, List<string>>();
    50       var postfix2infix = new Dictionary<string, string>();
     43      var hashToRowIdx = new Dictionary<string, int>();
     44      var hashToInfix = new Dictionary<string, string>();
    5145
    5246      // read all sentences and determine shortest sentences
    53       using (var reader = new StreamReader(allSentencesFileName)) {
     47      using (var reader = new StreamReader(
     48        new System.IO.Compression.GZipStream(
     49          new FileStream(sentencesFileName, FileMode.Open, FileAccess.Read),
     50          System.IO.Compression.CompressionMode.Decompress))) {
    5451        // read header
    5552        reader.ReadLine();
    5653        int nSentences = 0;
     54        var sw = new Stopwatch();
     55        sw.Start();
    5756        while (!reader.EndOfStream) {
    5857          var line = reader.ReadLine();
     58          nSentences++;
    5959          var toks = line.Split(';');
    6060          var hash = toks[0];
    6161          var length = toks[1];
    62           var postfix = toks[2];
     62          //var postfix = toks[2];
    6363          var infix = toks[3];
    64           List<string> alternativesList;
    65           if (!hash2Postfix.TryGetValue(hash, out alternativesList)) {
    66             alternativesList = new List<string>(1);
    67             hash2Postfix.Add(hash, alternativesList);
     64          string expr;
     65          if (!hashToInfix.TryGetValue(hash, out expr)) {
     66            hashToInfix.Add(hash, infix);
     67            hashToRowIdx.Add(hash, nSentences);
    6868          }
    69           alternativesList.Add(postfix);
    70           postfix2infix.Add(postfix, infix);
    71           nSentences++;
     69          else if(expr.Length > infix.Length) {
     70            hashToInfix[hash] = infix;  // keep only shortest
     71            hashToRowIdx[hash] = nSentences;
     72          }
     73          if (nSentences % PERF_STATS_UPDATE_INTERVAL == PERF_STATS_UPDATE_INTERVAL-1) {
     74            Console.WriteLine("Read perf: {0} sentences in {1}ms", PERF_STATS_UPDATE_INTERVAL, sw.ElapsedMilliseconds);
     75            sw.Restart();
     76          }
    7277        }
    7378
    74         Console.WriteLine("{0} {1}", nSentences, hash2Postfix.Count);
     79        Console.WriteLine("{0} {1}", nSentences, hashToInfix.Count);
    7580        //Evaluate(toks[1], xs, evalBuf);
    7681      }
    7782
    78       List<double[]> functions = new List<double[]>();
    79       List<string> sentences = new List<string>();
    80       List<double[]> qualities = new List<double[]>(); // we might have multiple target functions to which we might compare
     83      Scale(ys_keijzer4);
     84      Scale(ys_pagie);
    8185
    82       var ds = new Dataset(new string[] { "X" }, new IList[] { xs });
    83       foreach (var kvp in hash2Postfix) {
    84         var ls = kvp.Value;
    85         var sentence = FindShortest(ls);
    86         //EvaluatePostfix(sentence, xs, evalBuf);
    87         evalBuf = EvaluateInfix(postfix2infix[sentence], ds).ToArray();
    88         if (evalBuf.Any(ei => double.IsInfinity(ei) || double.IsNaN(ei))) {
    89           Console.WriteLine("skipping {0} {1}", evalBuf.Average(), sentence);
    90         } else {
    91           try {
    92             Scale(evalBuf);
    93             functions.Add((double[])evalBuf.Clone());
    94             sentences.Add(sentence);
    95             OnlineCalculatorError error;
    96             var r2_pagie = OnlinePearsonsRSquaredCalculator.Calculate(evalBuf, ys_pagie, out error);
    97             if (error != OnlineCalculatorError.None) r2_pagie = 0.0;
    98             var r2_keijzer4 = OnlinePearsonsRSquaredCalculator.Calculate(evalBuf, ys_keijzer4, out error);
    99             if (error != OnlineCalculatorError.None) r2_keijzer4 = 0.0;
    100             qualities.Add(new double[] { r2_pagie, r2_keijzer4});
    101           } catch (ArgumentException e) {
    102             // scaling failed
     86      // output all functions
     87      using (var writer = new StreamWriter(
     88        new System.IO.Compression.GZipStream(
     89          new FileStream(outputFileName, FileMode.OpenOrCreate),
     90          System.IO.Compression.CompressionMode.Compress))) {
     91        var sw = new Stopwatch();
     92        sw.Start();
     93
     94        var ds = new Dataset(new string[] { "X" }, new IList[] { xs });
     95        writer.WriteLine("{0};{1};{2};{3};{4};{5}", "Hash", "RowIdx (in allSentences)", "NMSE pagie", "NMSE keijzer4", "infix",
     96          string.Join(";", Enumerable.Range(0, xs.Length).Select(i => "eval" + i)));
     97        int nSentences = 0;
     98        foreach (var kvp in hashToInfix) {
     99          var hash = kvp.Key;
     100          var infixExpr = kvp.Value;
     101          evalBuf = EvaluateInfix(infixExpr, ds).ToArray();
     102          if (evalBuf.Any(ei => double.IsInfinity(ei) || double.IsNaN(ei))) {
     103            //Console.WriteLine("skipping {0} {1}", evalBuf.Average(), infixExpr);
     104            //Console.Write(".");
     105          } else {
     106            try {
     107              Scale(evalBuf);
     108              // functions.Add((double[])evalBuf.Clone());
     109              // sentences.Add(sentence);
     110              OnlineCalculatorError error;
     111              var nmse_pagie = OnlineNormalizedMeanSquaredErrorCalculator.Calculate(evalBuf, ys_pagie, out error);
     112              if (error != OnlineCalculatorError.None) nmse_pagie = 10;
     113              var nmse_keijzer = OnlineNormalizedMeanSquaredErrorCalculator.Calculate(evalBuf, ys_keijzer4, out error);
     114              if (error != OnlineCalculatorError.None) nmse_keijzer = 10;
     115              writer.WriteLine("{0};{1};{2};{3};{4};{5}", hash,  hashToRowIdx[hash], nmse_pagie, nmse_keijzer, infixExpr,
     116                string.Join(";", evalBuf.Select(fi => fi.ToString())));
     117            } catch (ArgumentException e) {
     118              // scaling failed
     119            }
     120          }
     121         
     122          if (nSentences++ % PERF_STATS_UPDATE_INTERVAL == PERF_STATS_UPDATE_INTERVAL-1) {
     123            Console.WriteLine("Eval perf: {0} sentences in {1}ms expected time remaining: {2}min",
     124              PERF_STATS_UPDATE_INTERVAL, sw.ElapsedMilliseconds,
     125              (hashToRowIdx.Count - nSentences) / (double)PERF_STATS_UPDATE_INTERVAL * sw.ElapsedMilliseconds / 1000 / 60);
     126            sw.Restart();
    103127          }
    104128        }
    105129      }
    106 
    107 
    108       List<int> clusters;
    109       List<double> distances;
    110       // DEACTIVATED FOR NOW -> USE LARGEVIS in R instead
    111       // Flann.FindClusters(functions, out clusters, out distances, 100);
    112       clusters = functions.Select(_ => 0).ToList();
    113       distances = functions.Select(_ => 0.0).ToList();
    114       //
    115       // output all clusters and functions
    116       using (var writer = new StreamWriter(new System.IO.Compression.GZipStream(new FileStream(outputFileName, FileMode.OpenOrCreate), System.IO.Compression.CompressionMode.Compress))) {
    117         for (int i = 0; i < functions.Count; i++) {
    118           writer.WriteLine("{0};{1};{2};{3};{4};{5}", clusters[i], distances[i], string.Join(";", qualities[i]), sentences[i], postfix2infix[sentences[i]], string.Join(";", functions[i].Select(fi => fi.ToString())));
    119         }
    120       }
    121       //
    122       // var funClusters = functions.Zip(clusters, (f, c) => Tuple.Create(f, c)).GroupBy(t => t.Item2);
    123       // var dtView = new DataTableView();
    124       // dtView.Size = new Size(800, 600);
    125       //
    126       // foreach (var funCluster in funClusters) {
    127       //   // draw the functions for each cluster into a separate png
    128       //   // var dtName = string.Format("R² {0}", Enumerable.Range(0, qualities.Count).Where(idx => clusters[idx] == funCluster.Key).Select(idx => qualities[idx]).Average());
    129       //   var dtName = "Cluster";
    130       //   var dt = new DataTable(dtName, dtName);
    131       //   var rows = new List<DataRow>();
    132       //   int i = 0;
    133       //   foreach (var fun in funCluster.Select(t => t.Item1)) {
    134       //     var name = i.ToString();
    135       //     var dr = new DataRow(name, name, fun);
    136       //     rows.Add(dr);
    137       //     i++;
    138       //   }
    139       //   dt.Rows.AddRange(rows);
    140       //   dtView.Content = dt;
    141       //   using (var bm = new Bitmap(800, 600)) {
    142       //     dtView.DrawToBitmap(bm, new Rectangle(0, 0, 800, 600));
    143       //     bm.Save(Path.Combine(clusterFolder, string.Format("cluster_{0,3}.png", funCluster.Key)));
    144       //   }
    145       // }
    146130    }
    147 
    148 
    149 
    150     private static string FindShortest(List<string> ls) {
    151       var minElem = ls.First();
    152       for (int i = 1; i < ls.Count; i++) {
    153         if (ls[i].Length < minElem.Length) minElem = ls[i];
    154       }
    155       return minElem;
    156     }
    157 
    158131
    159132
    160133    #region evaluation
    161134
    162     // scaling to zero-mean unit variance
     135    // scaling to zero-mean unit variance  (ignore NaN and +/-Inf.
    163136    private static void Scale(double[] evalBuf) {
    164137      double mean;
    165138      double variance;
    166       var max = evalBuf.Max();
     139      var max = evalBuf.Select(xi=>Math.Abs(xi)).Max();
    167140      for (int i = 0; i < evalBuf.Length; i++) {
    168141        evalBuf[i] /= max;
    … …  
    170143
    171144      OnlineCalculatorError error, varError;
    172       OnlineMeanAndVarianceCalculator.Calculate(evalBuf, out mean, out variance, out error, out varError);
     145      OnlineMeanAndVarianceCalculator.Calculate(evalBuf.Where(xi => !double.IsNaN(xi) && !double.IsInfinity(xi)), out mean, out variance, out error, out varError);
    173146      if(error!=OnlineCalculatorError.None || varError != OnlineCalculatorError.None) {
    174147        throw new ArgumentException("Cannot scale vector");
    … …  
    176149
    177150      for (int i = 0; i < evalBuf.Length; i++) {
     151        if (double.IsNaN(evalBuf[i])) evalBuf[i] = mean;
     152        else if (double.IsPositiveInfinity(evalBuf[i])) evalBuf[i] = 10;
     153        else if (double.IsNegativeInfinity(evalBuf[i])) evalBuf[i] = -10.0;
    178154        evalBuf[i] = 1.0 / variance * evalBuf[i] + mean;
    179155      }
    … …  
    202178      var interpreter = new SymbolicDataAnalysisExpressionTreeLinearInterpreter();
    203179      return interpreter.GetSymbolicExpressionTreeValues(tree, ds, Enumerable.Range(0, ds.Rows));
    204     }                                                 
    205 
    206     /*
    207     // evaluates postfix expressions (only for a very specific format)
    208     private static void EvaluatePostfix(string postfixExpr, double[] xs, double[] evalBuf) {
    209       int topOfStack = -1;
    210       Evaluate(postfixExpr, 0, xs, ref topOfStack);
    211       Array.Copy(stack[topOfStack], evalBuf, evalBuf.Length);
    212     }
    213 
    214    
    215     private static void Evaluate(string postfixExpr, int exprPos, double[] xs, ref int topOfStack) {
    216       while (exprPos < postfixExpr.Length) {
    217         switch (postfixExpr[exprPos]) {
    218           case '+': {
    219               exprPos += 2;
    220               var a = stack[topOfStack];
    221               var b = stack[topOfStack - 1];
    222               for (int i = 0; i < N; i++) {
    223                 b[i] += a[i];
    224               }
    225               topOfStack--;
    226               break;
    227             }
    228           case '*': {
    229               exprPos += 2;
    230               var a = stack[topOfStack];
    231               var b = stack[topOfStack - 1];
    232               for (int i = 0; i < N; i++) {
    233                 b[i] *= a[i];
    234               }
    235               topOfStack--;
    236               break;
    237             }
    238           case 'X': {
    239               exprPos += 2;
    240               topOfStack++;
    241               Array.Copy(xs, stack[topOfStack], N);
    242               break;
    243             }
    244           case 'c': {
    245               if (postfixExpr[exprPos + 1] == 'o') {
    246                 // cos
    247                 exprPos += 4;
    248                 var a = stack[topOfStack];
    249                 for (int i = 0; i < N; i++) {
    250                   a[i] = Math.Cos(a[i]);
    251                 }
    252                 break;
    253               } else {
    254                 exprPos += 2;
    255                 // put 1 onto top of stack     // BUG!
    256                 topOfStack++;
    257                 var a = stack[topOfStack];
    258                 for (int i = 0; i < N; i++) a[i] = 1.0;
    259                 break;
    260               }
    261             }
    262           case 's': {
    263               // sin
    264               exprPos += 4;
    265               var a = stack[topOfStack];
    266               for (int i = 0; i < N; i++) {
    267                 a[i] = Math.Sin(a[i]);
    268               }
    269               break;
    270             }
    271           case 'l': {
    272               // log
    273               exprPos += 4;
    274               var a = stack[topOfStack];
    275               for (int i = 0; i < N; i++) {
    276                 a[i] = Math.Log(a[i]);
    277               }
    278 
    279               break;
    280             }
    281           case 'e': {
    282               // exp
    283               exprPos += 4;
    284               var a = stack[topOfStack];
    285               for (int i = 0; i < N; i++) {
    286                 a[i] = Math.Exp(a[i]);
    287               }
    288 
    289               break;
    290             }
    291           case 'i': {
    292               // inv
    293               exprPos += 4;
    294               var a = stack[topOfStack];
    295               for (int i = 0; i < N; i++) {
    296                 a[i] = 1.0 / a[i];
    297               }
    298               break;
    299             }
    300           default: {
    301               throw new InvalidOperationException(string.Format("Cannot handle {0} in {1}", postfixExpr[exprPos], postfixExpr));
    302             }
    303         }
    304       }
    305     }
    306     */
     180    }         
    307181    #endregion
    308182  }

  • branches/2886_SymRegGrammarEnumeration/ExpressionClustering_R/ClusteringScript.R

    r15903 r15924  
    11library(largeVis)
    22library(ggplot2)
    3 
    4 eval7 <- read.csv2("c:/Users/P24581/Desktop/evaluations_2018-04-11_12-50_TreeSize-7.csv",header = TRUE, sep = ";", dec=",")
    5 eval7_qs <- eval7[,3]
    6 eval7_x <- eval7[,5:105]
    7 lv <- largeVis(eval7_x, dim=2, K = 50, n_trees = 50) # TODO scale?
    8 clusters <- hdbscan(lv, minPts = 3, K=30)
    9 
    10 m <- data.frame(x=t(lv$coords)[,1], y=t(lv$coords)[,2], c=clusters$clusters, q=eval7_qs)
     3library(dplyr)
    114
    125
    13 ggplot(data=m, aes(x=x, y=y)) + geom_point(aes(color=q))
     6
     7sentenceFileName <- "D:/heal/documents/trunk/Publications/2018/GPTP/data/evaluations_allSentences_2018-04-13_16-40_TreeSize-7_1d.csv.gz";
     8evalData <- read.csv(sentenceFileName,header = TRUE, sep = ";", dec=",");
     9qualities <- evalData$R2.keijzer4;
     10outputs <- evalData[,6:105];
     11
     12lv <- largeVis(outputs, dim=2, K = 50, n_trees = 50) # TODO scale?
     13clusters <- hdbscan(lv, minPts = 3, K=50);
     14
     15
     16
     17m <- data.frame(x=t(lv$coords)[,1], y=t(lv$coords)[,2], c=clusters$clusters, q=qualities, outputs)
     18# plot mapped points
     19ggplot(data=m, aes(x=x, y=y)) + geom_point(aes(color=c))  + theme(legend.position = "none") ;
     20
     21cluster_n <- dplyr::filter(m, c==5);
     22cluster_evals <- data.frame(x=seq(1,100,1), t(cluster_n[,5:104]))
     23evals_cluster_n <- tidyr::gather(cluster_evals,"f", "fx", 2:ncol(cluster_evals))
     24
     25p <- ggplot(evals_cluster_n, aes(x=x, y=fx,color=f)) + geom_line();
     26p
     27
Note: See TracChangeset for help on using the changeset viewer.