public static PCAResults PCA(String csvPath, int numberOfFactors)
{
try
{
REngine engine;
//init the R engine
engine = getInststance();
csvPath = csvPath.Replace('\\', '/');
//executing script
if (numberOfFactors > 0)
{
engine.Evaluate("library(psych)\n" +
"data <- read.csv('" + csvPath + "', sep = ';')\n" +
"pca <- principal(data, " + numberOfFactors + ", rotate='varimax')\n" +
"pcaTable <- as.data.frame.matrix(rbind(pca$values, pca$values/sum(pca$values)*100, cumsum(pca$values), cumsum(pca$values/sum(pca$values)*100)))\n" +
"loadings <- as.data.frame.matrix(t(pca$loadings))");
}
else
{
engine.Evaluate("library(psych)\n" +
"data <- read.csv('" + csvPath + "', sep = ';')\n" +
"pca <- principal(data, rotate='varimax')\n" +
"pcaTable <- as.data.frame.matrix(rbind(pca$values, pca$values/sum(pca$values)*100, cumsum(pca$values), cumsum(pca$values/sum(pca$values)*100)))\n" +
"loadings <- as.data.frame.matrix(t(pca$loadings))");
}
PCAResults pCAResults = new PCAResults();
pCAResults.pcaTable = engine.GetSymbol("pcaTable").AsDataFrame();
pCAResults.loadings = engine.GetSymbol("loadings").AsDataFrame();
Clear();
return(pCAResults);
}
catch (Exception e)
{
Console.WriteLine(e.StackTrace);
return(null);
}
}
private void buttonRunACP_Click(object sender, EventArgs e)
{
using (var form = new ACPForm(viewModel.getNumberOfQuestions()))
{
form.PropertyChanged += this.viewModel.Parallel;
var result = form.ShowDialog();
if (result == DialogResult.OK)
{
PCAResults results = this.viewModel.ACP(form.viewModel.factorNumber);
if (results != null)
{
dataGridPrincipalComponents.DataSource = DataTableManager.PCATableToDataTable(results.pcaTable);
dataGridPCALoadings.DataSource = DataTableManager.PCALoadingstoDataTable(results.loadings);
}
else
{
MessageBox.Show("An error has been occured or number of chosen factors is more than the number of questions", "Error", MessageBoxButtons.OK, MessageBoxIcon.Error);
}
factorisationResults.Visible = true;
}
}
}
/// <summary>
/// Calculates a matrix which transforms a set of signals to a the principal components of those signals.
/// </summary>
/// <param name="channels"></param>
/// <param name="length"></param>
/// <returns>Returns a matrix which transforms the channels (after being normalized) into the principal components</returns>
public static PCAResults CalculateMatrix(double[][] channels, int length)
{
var result = new PCAResults();
// Initilize Matrix and copy channels.
double[][] matrix = new double[channels.Length][];
double[][] channels2 = new double[channels.Length][];
for (int i = 0; i < channels.Length; i++)
{
matrix[i] = new double[channels.Length];
channels2[i] = channels[i].ToArray();
}
double[] sdSumByComponent = new double[channels.Length + 1]; // We add an extra "null component" to make our following code easier.
// Iterate to calculate it.
for (int componentIndex = 0; componentIndex < channels.Length; componentIndex++)
{
var p = InternalCalculateP(channels2, length);
matrix[componentIndex] = p.ToArray();
// Subtract the current component from the data.
double[] channelMeans = (from currentChannel in channels2 select currentChannel.Average()).ToArray();
double[] channelSDs = new double[channels.Length];
for (int channelIndex = 0; channelIndex < channels.Length; channelIndex++)
{
channelSDs[channelIndex] = Math.Pow((from sample in channels2[channelIndex] select Math.Pow((sample - channelMeans[channelIndex]), 2.0)).Average(), 0.5);
}
sdSumByComponent[componentIndex] = (from x in channelSDs select x * x).Sum();
result.MeansByStage.Add(channelMeans.ToArray());
result.StandardDeviationByStage.Add(channelSDs.ToArray());
for (int sampleIndex = 0; sampleIndex < length; sampleIndex++)
{
double dotProduct = 0.0;
for (int channelIndex = 0; channelIndex < channels.Length; channelIndex++)
{
var channel = channels2[channelIndex];
double sample = channel[sampleIndex];
// Normalize sample.
var sample_n = (sample - channelMeans[channelIndex]) / channelSDs[channelIndex];
dotProduct += sample_n * p[channelIndex];
channel[sampleIndex] = sample;
}
for (int channelIndex = 0; channelIndex < channels.Length; channelIndex++)
{
var channel = channels2[channelIndex];
double sample = channel[sampleIndex];
double sample_n = (sample - channelMeans[channelIndex]) / channelSDs[channelIndex];
// Subtract x*p.
sample_n -= p[channelIndex] * dotProduct;
double sample_p = (sample_n * channelSDs[channelIndex]) + channelMeans[channelIndex];
channel[sampleIndex] = sample_p;
}
}
}
double[] varExplainedByComponent = new double[channels.Length];
//Calculate the portion of variance explained by component.
for (int componentIndex = 0; componentIndex < channels.Length; componentIndex++)
{
varExplainedByComponent[componentIndex] = (sdSumByComponent[componentIndex] - sdSumByComponent[componentIndex + 1]) / sdSumByComponent[0];
}
result.Matrix = matrix;
result.VarianceExplainedByComponent = varExplainedByComponent;
return(result);
}
