public static void calcTreesVariance(List <int>[] table, double[][] label, int labelindex, string filename)
{
List <double> average = new List <double>();
List <double> averaged_variance = new List <double>();
//calc avg
for (int i = 0; i < table.Count(); i++)
{
double tmpAvg = 0;
for (int j = 0; j < table[i].Count(); j++)
{
tmpAvg += label[table[i][j]][labelindex];
}
average.Add(tmpAvg / table[i].Count());
}
//calc var
for (int i = 0; i < table.Count(); i++)
{
double tmpVar = 0;
for (int j = 0; j < table[i].Count(); j++)
{
tmpVar += (label[table[i][j]][labelindex] - average[i]) * (label[table[i][j]][labelindex] - average[i]);
}
averaged_variance.Add(tmpVar / table[i].Count());
}
PrintEngine.printList(averaged_variance, filename);
}
public DimReduction(double[][] trainingMatrix)
{
//Create the Principal Component Analysis
_pca = new ModifedPca(trainingMatrix);
_pca.Compute();
PrintEngine.printList(_pca.Eigenvalues.ToList(), Form1.MainFolderName + "eigvalues.txt");
}
public static void printForestProperties(List <GeoWave>[] rfTreeArr, string analysisFolderName)
{
if (Form1.u_config.saveTressCB != "1")
{
return;
}
if (!Directory.Exists(analysisFolderName + "\\archive"))
{
Directory.CreateDirectory(analysisFolderName + "\\archive");
}
for (int i = 0; i < rfTreeArr.Count(); i++)
{
PrintEngine.printWaveletsProperties(rfTreeArr[i], analysisFolderName + "\\archive\\waveletsPropertiesTree_" + i.ToString() + ".txt");
}
}
private void btnScript_Click(object sender, EventArgs e)
{
set2Config();
Refresh();
u_config.printConfig(@"C:\Wavelets decomposition\config.txt");
// AmazonS3Client client = Helpers.configAmazonS3ClientS3Client();
UseS3 = UseS3CB.Checked;
rumPrallel = rumPrallelCB.Checked;
runBoosting = runBoostingCB.Checked;
runProoning = runProoningCB.Checked;
runBoostingProoning = runBoostingProoningCB.Checked;
runRFProoning = runRFProoningCB.Checked;
runRf = runRfCB.Checked;
runBoostingLearningRate = runBoostingLearningRateCB.Checked;
bucketName = bucketTB.Text;
string results_path = @ResultsTB.Text;
string db_path = @DBTB.Text + "\\";//@"C:\Users\Administrator\Dropbox\ADA\ada_valid\"; //"D:\\Phd\\Shai\\code\\tests\\helix tests\\noise_5\\noise_5\\"; // "C:\\reasearch\\tests\\lena\\";
//get dir
MainFolderName = results_path;
//Helpers.createMainDirectoryOrResultPath(results_path, bucketName, client);
Helpers.createMainDirectoryOrResultPath(results_path, bucketName);
//READ DATA
DB db = new DB();
db.training_dt = db.getDataTable(db_path + "trainingData.txt");
db.testing_dt = db.getDataTable(db_path + "testingData.txt");
db.validation_dt = db.getDataTable(db_path + "ValidData.txt");
db.training_label = db.getDataTable(db_path + "trainingLabel.txt");
db.testing_label = db.getDataTable(db_path + "testingLabel.txt");
db.validation_label = db.getDataTable(db_path + "ValidLabel.txt");
upper_label = db.training_label.Max();
lower_label = db.training_label.Min();
double trainingPercent = double.Parse(trainingPercentTB.Text); // 0.02;
long rowToRemoveFrom = Convert.ToInt64(db.training_dt.Count() * trainingPercent);
db.training_dt = db.training_dt.Where((el, i) => i < rowToRemoveFrom).ToArray();
db.training_label = db.training_label.Where((el, i) => i < rowToRemoveFrom).ToArray();
db.testing_dt = db.testing_dt.Where((el, i) => i < rowToRemoveFrom).ToArray();
db.testing_label = db.testing_label.Where((el, i) => i < rowToRemoveFrom).ToArray();
db.validation_dt = db.training_dt.Where((el, i) => i < rowToRemoveFrom).ToArray();
db.validation_label = db.validation_label.Where((el, i) => i < rowToRemoveFrom).ToArray();
//REDUCE DIM, GLOBAL PCA
if (usePCA.Checked)
{
DimReduction dimreduction = new DimReduction(db.training_dt);
db.PCAtraining_dt = dimreduction.getGlobalPca(db.training_dt);
db.PCAtesting_dt = dimreduction.getGlobalPca(db.testing_dt);
db.PCAvalidation_dt = dimreduction.getGlobalPca(db.validation_dt);
}
else
{
//de-activate pca for dbg
db.PCAtraining_dt = db.training_dt;
db.PCAtesting_dt = db.testing_dt;
db.PCAvalidation_dt = db.validation_dt;
}
db.PCAtraining_GridIndex_dt = new long[db.PCAtraining_dt.Count()][];
for (int i = 0; i < db.PCAtraining_dt.Count(); i++)
{
db.PCAtraining_GridIndex_dt[i] = new long[db.PCAtraining_dt[i].Count()];
}
//BOUNDING BOX AND MAIN GRID
boundingBox = db.getboundingBox(db.PCAtraining_dt);
MainGrid = db.getMainGrid(db.PCAtraining_dt, boundingBox, ref db.PCAtraining_GridIndex_dt);
//READ CONFIG
methodConfig mc = new methodConfig(true);
int Nloops = int.Parse(NloopsTB.Text) - 1;
int Kfolds = 0;
if (int.TryParse(croosValidTB.Text, out Kfolds))
{
Nloops = Kfolds - 1;
}
for (int k = 0; k < Nloops; k++)
{
mc.boostlamda_0.Add(3.8); // - create variant in number of pixels
}
mc.generateRecordConfigArr();
for (int k = 0; k < mc.recArr.Count(); k++)
{
//manual set indeces of categorical variable
//mc.recArr[k].indOfCategorical.Add(3);
mc.recArr[k].learningType = (recordConfig.LearnigType)comboLearningType.SelectedIndex; // regression, binary class, multy class
if (mc.recArr[k].learningType == recordConfig.LearnigType.BinaryClassification)
{
mc.recArr[k].binaryMinClass = lower_label;
mc.recArr[k].binaryMaxClass = upper_label;
mc.recArr[k].midClassValue = 0.5 * (lower_label + upper_label);
}
mc.recArr[k].dim = NfeaturesTB.Text == @"all" ? db.PCAtraining_dt[0].Count() : int.Parse(evaluateString(NfeaturesTB.Text, k));
mc.recArr[k].approxThresh = double.Parse(evaluateString(approxThreshTB.Text, k)); // 0.1;
mc.recArr[k].partitionErrType = int.Parse(evaluateString(partitionTypeTB.Text, k)); //2;
mc.recArr[k].minWaveSize = int.Parse(evaluateString(minNodeSizeTB.Text, k)); //1;//CHANGE AFTER DBG
mc.recArr[k].hopping_size = int.Parse(evaluateString(waveletsSkipEstimationTB.Text, k)); //25;// 10 + 5 * (k + 1);// +5 * (k % 10);// 1;//25;
mc.recArr[k].test_error_size = double.Parse(evaluateString(waveletsPercentEstimationTB.Text, k)); // +0.05 * (k % 10);// 1;// 0.1;//percent of waves to check
mc.recArr[k].NskipsinKfunc = double.Parse(evaluateString(boostingKfuncPercentTB.Text, k)); // 0.0025;
mc.recArr[k].rfBaggingPercent = double.Parse(evaluateString(bagginPercentTB.Text, k)); // 0.6;
mc.recArr[k].rfNum = int.Parse(evaluateString(NrfTB.Text, k)); // k + 1;//10 + k*10;// 100 / (k + 46) * 2;// int.Parse(Math.Pow(10, k + 1).ToString());
mc.recArr[k].boostNum = int.Parse(evaluateString(NboostTB.Text, k)); // 10;
mc.recArr[k].boostProoning_0 = int.Parse(evaluateString(NfirstPruninginBoostingTB.Text, k)); //13
mc.recArr[k].boostlamda_0 = double.Parse(evaluateString(boostingLamda0TB.Text, k)); // 0.01 - (k + 1) * 0.001; //0.05;// 0.0801 + k * 0.001;// Math.Pow(0.1, k);// 0.22 + k*0.005;
mc.recArr[k].NwaveletsBoosting = int.Parse(evaluateString(NfirstwaveletsBoostingTB.Text, k)); // 4;// k + 1;
//mc.recArr[k].learningRate = 0;// 0.01;
mc.recArr[k].boostNumLearningRate = int.Parse(evaluateString(NboostingLearningRateTB.Text, k)); // 55;// 18;
mc.recArr[k].percent_training_db = trainingPercent;
mc.recArr[k].BoundLevel = int.Parse(evaluateString(boundLevelTB.Text, k)); //1024;
mc.recArr[k].NDimsinRF = NfeaturesrfTB.Text == @"all" ? db.PCAtraining_dt[0].Count() : int.Parse(evaluateString(NfeaturesrfTB.Text, k));
mc.recArr[k].split_type = int.Parse(evaluateString(splitTypeTB.Text, k)); //0
mc.recArr[k].NormLPType = int.Parse(evaluateString(errTypeEstimationTB.Text, k));
mc.recArr[k].RFpruningTestRange[1] = int.Parse(evaluateString(RFpruningEstimationRange1TB.Text, k)); // 12;// k + 9;
mc.recArr[k].boundDepthTree = int.Parse(evaluateString(boundDepthTB.Text, k)); //1024;
mc.recArr[k].CrossValidFold = k;
// 2m0rr0w2 save labels dim in confif
mc.recArr[k].labelDim = db.training_label[0].Count();
//mc.recArr[k].boostNum = t ;// tmp to delete !!!!!!!
//mc.recArr[k].RFwaveletsTestRange[0] = 25;
//mc.recArr[k].RFwaveletsTestRange[1] = 50;
}
// Helpers.createOutputDirectories(mc.recArr, client, u_config, bucketName, results_path);
Helpers.createOutputDirectories(mc.recArr, u_config, bucketName, results_path);
//SET ID ARRAY LIST
List <int> trainingID = Enumerable.Range(0, db.PCAtraining_dt.Count()).ToList();
List <int> testingID = Enumerable.Range(0, db.PCAtesting_dt.Count()).ToList();
//cross validation
List <List <int> > trainingFoldId = new List <List <int> >();
List <List <int> > testingFoldId = new List <List <int> >();
Random ran = new Random(2);
List <int> training_rand = trainingID.OrderBy(x => ran.Next()).ToList().GetRange(0, trainingID.Count);
//THE LARGEST GROUP IS TRAINING
if (int.TryParse(croosValidTB.Text, out Kfolds))
{
createCrossValid(Kfolds, training_rand, trainingFoldId, testingFoldId);
}
//bounding intervals
int[][] BB = new int[2][];
BB[0] = new int[boundingBox[0].Count()];
BB[1] = new int[boundingBox[0].Count()];
for (int i = 0; i < boundingBox[0].Count(); i++)
{
BB[1][i] = MainGrid[i].Count() - 1; //set last index in each dim
}
//loop over folds DEBUG i=2
for (int i = 3; i < mc.recArr.Count; i++)
{
recordConfig rc = mc.recArr[i];
Analizer analizer = new Analizer(MainFolderName + "\\" + rc.getShortName(), MainGrid, db, rc);
//exclude variables one by one
double[] numFeachuresVSerror = new double[rc.dim];
int indFeachureToExclude = -1;
double predResult = 0;
//loop over feachures
for (int j = rc.dim - 1; j >= 0; j--)
{
//DEBUG, DISABLE FEACHURE EXCLUDE
// j = 0;
if (!croosValidCB.Checked)
{
analizer.analize(trainingID, testingID, BB, ref indFeachureToExclude, ref predResult);
}
else
{
analizer.analize(trainingFoldId[i], testingFoldId[i], BB, ref indFeachureToExclude, ref predResult); //cross validation
}
numFeachuresVSerror[j] = predResult; // index j, used j+1 featchures
analizer.excludeFeatureFromDb(indFeachureToExclude);
}
PrintEngine.printBestErrorByNumberOfFeatchures(analizer._analysisFolderName, numFeachuresVSerror);
}
btnScript.BackColor = Color.Green;
}