public static void RocFromSortedMatches(List <PeakBoundsMatch> matches, NormalizeType normalizer, out PointPairList rocPoints)
{
double scalingFactor = GetScalingFactor(normalizer, matches);
rocPoints = new PointPairList();
int truePositives = 0;
int falsePositives = 0;
foreach (var match in matches)
{
if (match.IsFalsePositive)
{
falsePositives++;
}
if (match.IsMatch)
{
truePositives++;
}
// Null qValues get the worst score
double qValue = match.QValue ?? 1.0;
var rocPoint = new PointPair(falsePositives / (double)(truePositives + falsePositives), truePositives / scalingFactor)
{
Tag = qValue
};
rocPoints.Add(rocPoint);
}
}
protected override IParsedElement TryParseElement(List <string> lines, int index, out int end)
{
end = index + 1;
var text = lines[index].Trim();
while (text.EndsWith(","))
{
text = $"{text} {lines[end++].Trim()}";
}
var test = NormalizeType.Replace(text, string.Empty);
if (!FuncRegex.IsMatch(test) || VoidFuncRegex.IsMatch(test))
{
return(null);
}
var cppTypeMatch = FuncTypeRegex.Match(test);
if (!cppTypeMatch.Success)
{
return(null);
}
var nameMatch = FuncNameRegex.Match(test);
if (!nameMatch.Success)
{
return(null);
}
var paramsMatch = ParamsRegex.Match(test);
if (!paramsMatch.Success)
{
return(null);
}
var netType = cppTypeMatch.Value;
netType = GetKnownTypeOrDefault(netType);
var field = new ParsedFunc
{
Type = netType,
CppType = cppTypeMatch.Value,
Name = ToTitleCase(nameMatch.Value),
CppName = nameMatch.Value,
Params = TryParseParams(paramsMatch.Value),
CppParams = paramsMatch.Value,
//Comment = coment
};
return(field);
}
public void UpdateYAxis(ComboBox comboBox)
{
string selectedItem = (string)comboBox.SelectedItem;
zedGraphRoc.GraphPane.YAxis.Title.Text = selectedItem;
zedGraphFiles.GraphPane.YAxis.Title.Text = selectedItem;
Normalizer = (selectedItem == Resources.ComparePeakPickingDlg_ComparePeakPickingDlg_Total_Correct_Peaks) ? NormalizeType.total :
(selectedItem == Resources.ComparePeakPickingDlg_ComparePeakPickingDlg_Fraction_of_Manual_ID_s) ? NormalizeType.frac_manual
: NormalizeType.frac_all;
comboBoxYAxis.SelectedItem = selectedItem;
comboBoxFilesYAxis.SelectedItem = selectedItem;
}
public static void MakeRocLists(ComparePeakBoundaries comparer, NormalizeType normalizer, out PointPairList rocPoints)
{
var matches = comparer.Matches;
if (comparer.HasNoScores)
{
matches.Sort(PeakBoundsMatch.CompareQValue);
}
else
{
matches.Sort(PeakBoundsMatch.CompareScore);
}
RocFromSortedMatches(matches, normalizer, out rocPoints);
}
/// <summary>
/// Convert normalize type to the xgboost parameter string.
/// </summary>
/// <param name="type"></param>
/// <returns></returns>
public static string ToXGBoostString(this NormalizeType type)
{
switch (type)
{
case NormalizeType.Tree:
return("tree");
case NormalizeType.Forest:
return("forest");
default:
throw new ArgumentException("Unknown normalize type: " + type);
}
}
/// <summary>
/// Converts radians to degrees.
/// </summary>
/// <param name="radians"> Angle in radians. </param>
/// <param name="norm">
/// Indicate whether we should normalize the angle, and if so then how.
/// </param>
/// <returns> Angle in degrees. </returns>
public static float Degree(float radians, NormalizeType norm)
{
float degrees = (float)(radians * (180 / Math.PI));
switch (norm)
{
case NormalizeType.Normalize180:
while (degrees <= -180) degrees += 360;
while (degrees > 180) degrees -= 360;
break;
case NormalizeType.Normalize360:
while (degrees <= 0) degrees += 360;
while (degrees > 360) degrees -= 360;
break;
}
return degrees;
}
/// <summary>
/// Normalizes the specified value.
/// </summary>
/// <param name="value">The value.</param>
/// <param name="normalizeType">The normalize type.</param>
/// <returns></returns>
public static string Normalize(string value, NormalizeType normalizeType)
{
if (string.IsNullOrWhiteSpace(value))
{
return(value);
}
if (normalizeType == NormalizeType.CamelCase)
{
return(Char.ToLowerInvariant(value[0]) + value.Substring(1));
}
else if (normalizeType == NormalizeType.PascalCase)
{
return(Char.ToUpperInvariant(value[0]) + value.Substring(1));
}
return(value);
}
public static double GetScalingFactor(NormalizeType normalizer, IList <PeakBoundsMatch> matches)
{
if (normalizer == NormalizeType.total)
{
return(1);
}
else if (normalizer == NormalizeType.frac_manual)
{
return(matches.Count(match => !match.IsMissingTruePeak));
}
else if (normalizer == NormalizeType.frac_all)
{
return(matches.Count);
}
else
{
throw new InvalidDataException("Unrecognized y axis scaling option"); // Not L10N
}
}
public RenderTexture IDFT(NormalizeType normalizeType = NormalizeType.IDFT)
{
FFTUWithRadix(FFTType.IDFT, 32);
FFTVWithRadix(FFTType.IDFT, 32);
if (normalizeType == NormalizeType.IDFT)
{
Normalize(1.0f / n);
}
else if (normalizeType == NormalizeType.SYMMETRIC)
{
Normalize(1.0f / Mathf.Sqrt(n));
}
else
{
SwapTex();
}
return(fftTextureOut_);
}
public NoiseMapProperties(
long seed,
int sizeBound,
Vector2 offset,
float scale,
float chunkScale,
int octaves,
float lacunarity,
float persistance,
NormalizeType normalizeType)
{
this.seed = seed;
this.sizeBound = sizeBound;
this.offset = offset;
this.scale = scale;
this.chunkScale = chunkScale;
this.octaves = octaves;
this.lacunarity = lacunarity;
this.persistance = persistance;
this.normalizeType = normalizeType;
}
public static void RocFromSortedMatches(List <PeakBoundsMatch> matches, NormalizeType normalizer, out PointPairList rocPoints)
{
double scalingFactor = GetScalingFactor(normalizer, matches);
rocPoints = new PointPairList();
int truePositives = 0;
int falsePositives = 0;
foreach (var match in matches)
{
if (match.IsFalsePositive)
{
falsePositives++;
}
if (match.IsPickedApexBetweenCuratedBoundaries)
{
truePositives++;
}
var rocPoint = new PointPair(falsePositives / (double)(truePositives + falsePositives), truePositives / scalingFactor);
rocPoints.Add(rocPoint);
}
}
/// <summary>
/// Converts radians to degrees.
/// </summary>
/// <param name="radians"> Angle in radians. </param>
/// <param name="norm">
/// Indicate whether we should normalize the angle, and if so then how.
/// </param>
/// <returns> Angle in degrees. </returns>
public static double Degree(double radians, NormalizeType norm)
{
double degrees = radians * (180 / Math.PI);
switch (norm)
{
case NormalizeType.Normalize180:
while (degrees <= -180) degrees += 360;
while (degrees > 180) degrees -= 360;
break;
case NormalizeType.Normalize360:
while (degrees <= 0) degrees += 360;
while (degrees > 360) degrees -= 360;
break;
}
return degrees;
}
/// <summary>
/// Converts degrees to radians.
/// </summary>
/// <param name="degrees"> Angle in degrees. </param>
/// <param name="norm">
/// Indicate whether we should normalize the angle, and if so then how.
/// </param>
/// <returns> Angle in radians. </returns>
public static float Radian(float degrees, NormalizeType norm)
{
double radians = degrees * Math.PI / 180;
switch (norm)
{
case NormalizeType.Normalize180:
while (radians <= -Math.PI) radians += MathHelpers.PI2;
while (radians > Math.PI) radians -= MathHelpers.PI2;
break;
case NormalizeType.Normalize360:
while (radians <= 0) radians += MathHelpers.PI2;
while (radians > MathHelpers.PI2) radians -= MathHelpers.PI2;
break;
}
return (float)radians;
}
public static int[,] Normalize(Complex[,] Output, NormalizeType normalizeType)
{
int Width = Output.GetLength(0);
int Height = Output.GetLength(1);
double[,] FourierDouble = new double[Width, Height];
double[,] FourierLogDouble = new double[Width, Height];
int[,] FourierNormalizedInteger = new int[Width, Height];
double max = 0;
if (normalizeType == NormalizeType.Magnitude)
{
for (int i = 0; i <= Width - 1; i++)
{
for (int j = 0; j <= Height - 1; j++)
{
FourierDouble[i, j] = Output[i, j].Magnitude;
FourierLogDouble[i, j] = (double)Math.Log(1 + FourierDouble[i, j]);
}
}
max = FourierLogDouble[0, 0];
}
else
{
for (int i = 0; i <= Width - 1; i++)
{
for (int j = 0; j <= Height - 1; j++)
{
FourierDouble[i, j] = Output[i, j].Phase;
FourierLogDouble[i, j] = (double)Math.Log(1 + Math.Abs(FourierDouble[i, j]));
}
}
FourierLogDouble[0, 0] = 0;
max = FourierLogDouble[1, 1];
}
for (int i = 0; i <= Width - 1; i++)
{
for (int j = 0; j <= Height - 1; j++)
{
if (FourierLogDouble[i, j] > max)
{
max = FourierLogDouble[i, j];
}
}
}
for (int i = 0; i <= Width - 1; i++)
{
for (int j = 0; j <= Height - 1; j++)
{
FourierLogDouble[i, j] = FourierLogDouble[i, j] / max;
}
}
if (normalizeType == NormalizeType.Magnitude)
{
for (int i = 0; i <= Width - 1; i++)
{
for (int j = 0; j <= Height - 1; j++)
{
FourierNormalizedInteger[i, j] = (int)(2000 * FourierLogDouble[i, j]);
}
}
}
else
{
for (int i = 0; i <= Width - 1; i++)
{
for (int j = 0; j <= Height - 1; j++)
{
FourierNormalizedInteger[i, j] = (int)(255 * FourierLogDouble[i, j]);
}
}
}
return(FourierNormalizedInteger);
}
/// <summary>
/// Regression learner for XGBoost
/// </summary>
/// <param name="maximumTreeDepth">Maximum tree depth for base learners. (default is 3)</param>
/// <param name="learningRate">Boosting learning rate (xgb's "eta"). 0 indicates no limit. (default is 0.1)</param>
/// <param name="estimators">Number of estimators to fit. (default is 100)</param>
/// <param name="silent">Whether to print messages while running boosting. (default is false)</param>
/// <param name="objective">Specify the learning task and the corresponding learning objective. (default is LinearRegression)</param>
/// <param name="boosterType"> which booster to use, can be gbtree, gblinear or dart.
/// gbtree and dart use tree based model while gblinear uses linear function (default is gbtree)</param>
/// <param name="treeMethod">The tree construction algorithm used in XGBoost. See reference paper: https://arxiv.org/abs/1603.02754. (default is auto)</param>
/// <param name="samplerType">Type of sampling algorithm for DART. (default is uniform)</param>
/// <param name="normalizeType">Type of normalization algorithm for DART. (default is tree)</param>
/// <param name="dropoutRate">Dropout rate for DART (a fraction of previous trees to drop during the dropout). (default is 0.0)</param>
/// <param name="oneDrop">When this is true, at least one tree is always dropped during the dropout.
/// Allows Binomial-plus-one or epsilon-dropout from the original DART paper. (default is false)</param>
/// <param name="skipDrop">Probability of skipping the dropout procedure during a boosting iteration. (default is 0.0)
/// If a dropout is skipped, new trees are added in the same manner as gbtree.
/// Note that non-zero skip_drop has higher priority than rate_drop or one_drop.</param>
/// <param name="numberOfThreads">Number of parallel threads used to run xgboost. -1 means use all thread available. (default is -1)</param>
/// <param name="gamma">Minimum loss reduction required to make a further partition on a leaf node of the tree. (default is 0) </param>
/// <param name="minChildWeight">Minimum sum of instance weight(Hessian) needed in a child. (default is 1)</param>
/// <param name="maxDeltaStep">Maximum delta step we allow each tree's weight estimation to be. (default is 0)</param>
/// <param name="subSample">Subsample ratio of the training instance. (default is 1)</param>
/// <param name="colSampleByTree">Subsample ratio of columns when constructing each tree. (default is 1)</param>
/// <param name="colSampleByLevel">Subsample ratio of columns for each split, in each level. (default is 1)</param>
/// <param name="l1Regularization">L1 regularization term on weights. Also known as RegAlpha. (default is 0)</param>
/// <param name="l2Reguralization">L2 regularization term on weights. Also known as regLambda. (default is 1)</param>
/// <param name="scalePosWeight">Balancing of positive and negative weights. (default is 1)</param>
/// <param name="baseScore">The initial prediction score of all instances, global bias. (default is 0.5)</param>
/// <param name="seed">Random number seed. (default is 0)</param>
/// <param name="missing">Value in the data which needs to be present as a missing value. (default is NaN)</param>
public RegressionXGBoostLearner(
int maximumTreeDepth = 3,
double learningRate = 0.1,
int estimators = 100,
bool silent = true,
RegressionObjective objective = RegressionObjective.LinearRegression,
BoosterType boosterType = BoosterType.GBTree,
TreeMethod treeMethod = TreeMethod.Auto,
SamplerType samplerType = SamplerType.Uniform,
NormalizeType normalizeType = NormalizeType.Tree,
double dropoutRate = 0.0,
bool oneDrop = false,
double skipDrop = 0.0,
int numberOfThreads = -1,
double gamma = 0,
int minChildWeight = 1,
int maxDeltaStep = 0,
double subSample = 1,
double colSampleByTree = 1,
double colSampleByLevel = 1,
double l1Regularization = 0,
double l2Reguralization = 1,
double scalePosWeight = 1,
double baseScore = 0.5,
int seed = 0,
double missing = double.NaN)
{
ArgumentChecks.ThrowOnArgumentLessThan(nameof(maximumTreeDepth), maximumTreeDepth, 0);
ArgumentChecks.ThrowOnArgumentLessThanOrHigherThan(nameof(learningRate), learningRate, 0, 1.0);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(estimators), estimators, 1);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(numberOfThreads), numberOfThreads, -1);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(gamma), gamma, 0);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(minChildWeight), minChildWeight, 0);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(maxDeltaStep), maxDeltaStep, 0);
ArgumentChecks.ThrowOnArgumentLessThanOrHigherThan(nameof(subSample), subSample, 0, 1.0);
ArgumentChecks.ThrowOnArgumentLessThanOrHigherThan(nameof(colSampleByTree), colSampleByTree, 0, 1.0);
ArgumentChecks.ThrowOnArgumentLessThanOrHigherThan(nameof(colSampleByLevel), colSampleByLevel, 0, 1.0);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(l1Regularization), l1Regularization, 0);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(l2Reguralization), l2Reguralization, 0);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(scalePosWeight), scalePosWeight, 0);
m_parameters[ParameterNames.MaxDepth] = maximumTreeDepth;
m_parameters[ParameterNames.LearningRate] = (float)learningRate;
m_parameters[ParameterNames.Estimators] = estimators;
m_parameters[ParameterNames.Silent] = silent;
m_parameters[ParameterNames.objective] = objective.ToXGBoostString();
m_parameters[ParameterNames.Threads] = numberOfThreads;
m_parameters[ParameterNames.Gamma] = (float)gamma;
m_parameters[ParameterNames.MinChildWeight] = minChildWeight;
m_parameters[ParameterNames.MaxDeltaStep] = maxDeltaStep;
m_parameters[ParameterNames.SubSample] = (float)subSample;
m_parameters[ParameterNames.ColSampleByTree] = (float)colSampleByTree;
m_parameters[ParameterNames.ColSampleByLevel] = (float)colSampleByLevel;
m_parameters[ParameterNames.RegAlpha] = (float)l1Regularization;
m_parameters[ParameterNames.RegLambda] = (float)l2Reguralization;
m_parameters[ParameterNames.ScalePosWeight] = (float)scalePosWeight;
m_parameters[ParameterNames.BaseScore] = (float)baseScore;
m_parameters[ParameterNames.Seed] = seed;
m_parameters[ParameterNames.Missing] = (float)missing;
m_parameters[ParameterNames.ExistingBooster] = null;
m_parameters[ParameterNames.Booster] = boosterType.ToXGBoostString();
m_parameters[ParameterNames.TreeMethod] = treeMethod.ToXGBoostString();
m_parameters[ParameterNames.SampleType] = samplerType.ToXGBoostString();
m_parameters[ParameterNames.NormalizeType] = normalizeType.ToXGBoostString();
m_parameters[ParameterNames.RateDrop] = (float)dropoutRate;
m_parameters[ParameterNames.OneDrop] = oneDrop ? 1 : 0;
m_parameters[ParameterNames.SkipDrop] = (float)skipDrop;
}
/// <summary>
/// Classification learner for XGBoost. For classification problems,
/// XGBoost requires that target values are sequntial and start at 0.
/// </summary>
/// <param name="maximumTreeDepth">Maximum tree depth for base learners. (default is 3)</param>
/// <param name="learningRate">Boosting learning rate (xgb's "eta"). 0 indicates no limit. (default is 0.1)</param>
/// <param name="estimators">Number of estimators to fit. (default is 100)</param>
/// <param name="silent">Whether to print messages while running boosting. (default is false)</param>
/// <param name="objective">Specify the learning task and the corresponding learning objective. (default is softmax)</param>
/// <param name="boosterType"> which booster to use, can be gbtree, gblinear or dart.
/// gbtree and dart use tree based model while gblinear uses linear function (default is gbtree)</param>
/// <param name="treeMethod">The tree construction algorithm used in XGBoost. See reference paper: https://arxiv.org/abs/1603.02754. (default is auto)</param>
/// <param name="samplerType">Type of sampling algorithm for DART. (default is uniform)</param>
/// <param name="normalizeType">Type of normalization algorithm for DART. (default is tree)</param>
/// <param name="dropoutRate">Dropout rate for DART (a fraction of previous trees to drop during the dropout). (default is 0.0)</param>
/// <param name="oneDrop">When this is true, at least one tree is always dropped during the dropout.
/// Allows Binomial-plus-one or epsilon-dropout from the original DART paper. (default is false)</param>
/// <param name="skipDrop">Probability of skipping the dropout procedure during a boosting iteration. (default is 0.0)
/// If a dropout is skipped, new trees are added in the same manner as gbtree.
/// Note that non-zero skip_drop has higher priority than rate_drop or one_drop.</param>
/// <param name="numberOfThreads">Number of parallel threads used to run xgboost. -1 means use all thread avialable. (default is -1)</param>
/// <param name="gamma">Minimum loss reduction required to make a further partition on a leaf node of the tree. (default is 0) </param>
/// <param name="minChildWeight">Minimum sum of instance weight(hessian) needed in a child. (default is 1)</param>
/// <param name="maxDeltaStep">Maximum delta step we allow each tree's weight estimation to be. (default is 0)</param>
/// <param name="subSample">Subsample ratio of the training instance. (default is 1)</param>
/// <param name="colSampleByTree">Subsample ratio of columns when constructing each tree. (defualt is 1)</param>
/// <param name="colSampleByLevel">Subsample ratio of columns for each split, in each level. (defualt is 1)</param>
/// <param name="l1Regularization">L1 regularization term on weights. Also known as RegAlpha. (default is 0)</param>
/// <param name="l2Reguralization">L2 regularization term on weights. Also known as regLambda. (default is 1)</param>
/// <param name="scalePosWeight">Balancing of positive and negative weights. (default is 1)</param>
/// <param name="baseScore">The initial prediction score of all instances, global bias. (default is 0.5)</param>
/// <param name="seed">Random number seed. (defaukt is 0)</param>
/// <param name="missing">Value in the data which needs to be present as a missing value. (default is NaN)</param>
public ClassificationXGBoostLearner(int maximumTreeDepth = 3, double learningRate = 0.1, int estimators = 100,
bool silent = true,
ClassificationObjective objective = ClassificationObjective.Softmax,
BoosterType boosterType = BoosterType.GBTree,
TreeMethod treeMethod = TreeMethod.Auto,
SamplerType samplerType = SamplerType.Uniform,
NormalizeType normalizeType = NormalizeType.Tree,
double dropoutRate = 0.0,
bool oneDrop = false,
double skipDrop = 0.0,
int numberOfThreads = -1, double gamma = 0, int minChildWeight = 1,
int maxDeltaStep = 0, double subSample = 1, double colSampleByTree = 1,
double colSampleByLevel = 1, double l1Regularization = 0, double l2Reguralization = 1,
double scalePosWeight = 1, double baseScore = 0.5, int seed = 0,
double missing = double.NaN)
{
ArgumentChecks.ThrowOnArgumentLessThan(nameof(maximumTreeDepth), maximumTreeDepth, 0);
ArgumentChecks.ThrowOnArgumentLessThanOrHigherThan(nameof(learningRate), learningRate, 0, 1.0);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(estimators), estimators, 1);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(numberOfThreads), numberOfThreads, -1);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(gamma), gamma, 0);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(minChildWeight), minChildWeight, 0);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(maxDeltaStep), maxDeltaStep, 0);
ArgumentChecks.ThrowOnArgumentLessThanOrHigherThan(nameof(subSample), subSample, 0, 1.0);
ArgumentChecks.ThrowOnArgumentLessThanOrHigherThan(nameof(colSampleByTree), colSampleByTree, 0, 1.0);
ArgumentChecks.ThrowOnArgumentLessThanOrHigherThan(nameof(colSampleByLevel), colSampleByLevel, 0, 1.0);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(l1Regularization), l1Regularization, 0);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(l2Reguralization), l2Reguralization, 0);
ArgumentChecks.ThrowOnArgumentLessThan(nameof(scalePosWeight), scalePosWeight, 0);
m_parameters[ParameterNames.MaxDepth] = maximumTreeDepth;
m_parameters[ParameterNames.LearningRate] = (float)learningRate;
m_parameters[ParameterNames.Estimators] = estimators;
m_parameters[ParameterNames.Silent] = silent;
if (objective == ClassificationObjective.Softmax)
{
// SoftMax and SoftProp are the same objective,
// but softprop returns probabilities.
// So in order to always support PredictProbability,
// always use softprop for multi-class.
// Conversions to class labels is handled in the
// ClassificationXGBoostModel.
objective = ClassificationObjective.SoftProb;
}
m_parameters[ParameterNames.objective] = objective.ToXGBoostString();
m_parameters[ParameterNames.Threads] = numberOfThreads;
m_parameters[ParameterNames.Gamma] = (float)gamma;
m_parameters[ParameterNames.MinChildWeight] = minChildWeight;
m_parameters[ParameterNames.MaxDeltaStep] = maxDeltaStep;
m_parameters[ParameterNames.SubSample] = (float)subSample;
m_parameters[ParameterNames.ColSampleByTree] = (float)colSampleByTree;
m_parameters[ParameterNames.ColSampleByLevel] = (float)colSampleByLevel;
m_parameters[ParameterNames.RegAlpha] = (float)l1Regularization;
m_parameters[ParameterNames.RegLambda] = (float)l2Reguralization;
m_parameters[ParameterNames.ScalePosWeight] = (float)scalePosWeight;
m_parameters[ParameterNames.BaseScore] = (float)baseScore;
m_parameters[ParameterNames.Seed] = seed;
m_parameters[ParameterNames.Missing] = (float)missing;
m_parameters[ParameterNames.ExistingBooster] = null;
m_parameters[ParameterNames.Booster] = boosterType.ToXGBoostString();
m_parameters[ParameterNames.TreeMethod] = treeMethod.ToXGBoostString();
m_parameters[ParameterNames.SampleType] = samplerType.ToXGBoostString();
m_parameters[ParameterNames.NormalizeType] = normalizeType.ToXGBoostString();
m_parameters[ParameterNames.RateDrop] = (float)dropoutRate;
m_parameters[ParameterNames.OneDrop] = oneDrop ? 1 : 0;
m_parameters[ParameterNames.SkipDrop] = (float)skipDrop;
}
