/// <summary>
/// Evaluates the cross validation error for each fold.
/// </summary>
/// <param name="filePath">The training file path (after feature extraction).</param>
/// <param name="learningParameters">The learning parameters.</param>
/// <param name="nFolds">The number of folds.</param>
/// <param name="distance">The distance function to evaluate the scores.</param>
/// <returns>An array, each element containing the error over the fold.</returns>
public double[] CrossValidationScore(string filePath, LearningParameters learningParameters, int nFolds, DistanceFunction distance)
{
double[] perFoldError = new double[nFolds];
int currentLine = 0;
foreach (Tuple<WeightedPoint, List<string>> line in YieldLines(filePath, true, learningParameters))
{
currentLine++;
for (int i = 0; i < nFolds; i++)
if (currentLine % nFolds == i) // predict over the i-th field
perFoldError[i] += distance(line.Item1, PredictFromLine(_vectorsCV[i], line.Item2, learningParameters));
}
for (int i = 0; i < nFolds; i++)
perFoldError[i] = perFoldError[i] / currentLine * 5;
return perFoldError;
}
/// <summary>
/// Learns over each fold.
/// </summary>
/// <param name="filePath">The training file path (after feature extraction).</param>
/// <param name="nFolds">The number of folds.</param>
/// <param name="learningParameters">The learning parameters.</param>
public void CrossLearning(string filePath, int nFolds, LearningParameters learningParameters)
{
_vectorsCV = new List<Dictionary<string, StreamingCloud>>(nFolds);
for (int i = 0; i < nFolds; i++)
_vectorsCV.Add(new Dictionary<string, StreamingCloud>());
int currentLine = 0;
foreach (Tuple<WeightedPoint, List<string>> line in YieldLines(filePath, true, learningParameters))
{
currentLine++;
for (int i = 0; i < nFolds; i++)
if (currentLine % nFolds != i) // do not learn over the i-th field
foreach (string predictor in line.Item2)
_vectorsCV[i].TryAdd(predictor, line.Item1);
}
}
/// <summary>
/// Trains the model.
/// </summary>
/// <param name="filePath">The training file path (after feature extraction).</param>
/// <param name="learningParameters">The learning parameters.</param>
public void Train(string filePath, LearningParameters learningParameters)
{
foreach (Tuple<WeightedPoint, List<string>> line in YieldLines(filePath, true, learningParameters))
foreach (string predictor in line.Item2)
_vector.TryAdd(predictor, line.Item1);
string[] learned = _vector.Select(kvp => kvp.Key + "," + kvp.Value.ToString()).ToArray();
File.WriteAllLines(Environment.GetFolderPath(Environment.SpecialFolder.MyDocuments) + "\\Learned.txt", learned);
}
/// <summary>
/// Yields the lines from a file as predictors and final point (if train set)
/// </summary>
/// <param name="filePath">The training file path (after feature extraction).</param>
/// <param name="train">Set to true for train files, to return the final points as labels.</param>
/// <param name="learningParameters">The learning parameters.</param>
/// <returns></returns>
private IEnumerable<Tuple<WeightedPoint, List<string>>> YieldLines(string filePath, bool train, LearningParameters learningParameters)
{
using (StreamReader reader = new StreamReader(filePath))
{
string line = ""; //No header in feature files
while ((line = reader.ReadLine()) != null)
{
List<string> arrayLine = line.Split(',').ToList();
WeightedPoint bv = new WeightedPoint(0, 0, 0);
if (train)
{
string[] bidimVector = arrayLine[0].Split('_');
bv = new WeightedPoint(Convert.ToDouble(bidimVector[0], CultureInfo.GetCultureInfo("en-US")),
Convert.ToDouble(bidimVector[1], CultureInfo.GetCultureInfo("en-US")), 1);
arrayLine.RemoveAt(0);
}
arrayLine = arrayLine.CartesianProduct(learningParameters.Keyword);
yield return new Tuple<WeightedPoint, List<string>>(bv, arrayLine);
}
}
}
private WeightedPoint PredictFromLine(Dictionary<string, StreamingCloud> clouds,
List<string> predictors, LearningParameters predictionParameters)
{
List<WeightedPoint> bvs = new List<WeightedPoint>();
foreach (string predictor in predictors)
if (clouds.ContainsKey(predictor) && clouds[predictor].Size > predictionParameters.MinOccurences && clouds[predictor].Size < predictionParameters.MaxOccurences)
{
StreamingCloud cloud = clouds[predictor];
WeightedPoint wp = new WeightedPoint(cloud.Barycenter);
wp.Weight =
Math.Pow(cloud.Size, predictionParameters.SizeExponent) / Math.Pow(cloud.Dispersion, predictionParameters.DispersionExponent);
bvs.Add(wp);
}
if (bvs.Count == 0) // could not find any predictor respecting the learning conditions
return new WeightedPoint(-8.611317, 41.146504, 1);
return WeightedPoint.Barycenter(bvs);
}
private List<WeightedPoint> Predict(string filePath, LearningParameters learningParameters)
{
List<WeightedPoint> res = new List<WeightedPoint>();
foreach (Tuple<WeightedPoint, List<string>> line in YieldLines(filePath, false, learningParameters))
res.Add(PredictFromLine(_vector, line.Item2, learningParameters));
return res;
}
/// <summary>
/// Trains the model and generate the predictions.
/// </summary>
/// <param name="trainFilePath">The training file path (after feature extraction).</param>
/// <param name="testFilePath">The testing file path (after feature extraction).</param>
/// <param name="outFilePath">The predicted values file path.</param>
/// <param name="sampleSubmission">The sample submission file path (as provided by Kaggle).</param>
/// <param name="learningParameters">The learning parameters.</param>
public void TrainPredictAndWrite(string trainFilePath, string testFilePath, string outFilePath, string sampleSubmission,
LearningParameters learningParameters)
{
Train(trainFilePath, learningParameters);
List<WeightedPoint> predicted = Predict(testFilePath, learningParameters);
using (StreamReader reader = new StreamReader(sampleSubmission))
using (StreamWriter writer = new StreamWriter(outFilePath))
{
string line = reader.ReadLine();
writer.Write(line); // copy header
int i = 0;
while ((line = reader.ReadLine()) != null)
{
string idLine = line.Split(',')[0];
string res = idLine + ',' + predicted[i].ToString();
i++;
writer.Write(Environment.NewLine + res);
}
}
}
private void button4_Click(object sender, EventArgs e)
{
MessageBox.Show("Train file");
OpenFileDialog of = new OpenFileDialog();
of.ShowDialog();
if (!of.CheckFileExists) return;
string trainFilePath = of.FileNames[0];
StreamingLearning.DistanceFunction distance = Distances.Haversine;
string[] keywords = keywordsTbx.Text.Split(';'),
maxOccurences = maxoccurencesTbx.Text.Split(';'),
minOccurences = minOccurencesTbx.Text.Split(';'),
exps1 = expo1Tbx.Text.Split(';'),
exps2 = expo2Tbx.Text.Split(';');
string errors = "";
foreach (string keyword in keywords)
{
int nFolds = 5;
StreamingLearning sl = new StreamingLearning();
LearningParameters learningParameters = new LearningParameters(keyword + "_0_0_0_0");
sl.CrossLearning(trainFilePath, nFolds, learningParameters);
foreach (string maxOc in maxOccurences)
foreach (string minOc in minOccurences)
foreach (string exp1 in exps1)
foreach (string exp2 in exps2)
{
string learning = keyword + "_" + minOc + "_" + maxOc + "_" + exp1 + "_" + exp2;
learningParameters = new LearningParameters(learning);
double[] err = sl.CrossValidationScore(trainFilePath, learningParameters, nFolds, distance);
errors += learning + ";" + String.Join(";", err) + Environment.NewLine;
}
}
string fileName = Path.GetFileNameWithoutExtension(trainFilePath);
File.WriteAllText(Environment.GetFolderPath(Environment.SpecialFolder.MyDocuments) + "\\" + fileName + "_CrossValidation_1.1.csv", errors);
MessageBox.Show(errors);
}
private void button3_Click(object sender, EventArgs e)
{
MessageBox.Show("Test file");
OpenFileDialog of = new OpenFileDialog();
of.ShowDialog();
string testFilePath = of.FileName;
MessageBox.Show("Train file");
of.ShowDialog();
string trainFilePath = of.FileName;
MessageBox.Show("Sample submission");
of.ShowDialog();
string sample = of.FileName;
string[] keywords = keywordsTbx.Text.Split(';');
string[] maxOccurences = maxoccurencesTbx.Text.Split(';');
string[] minOccurences = minOccurencesTbx.Text.Split(';');
string[] exps1 = expo1Tbx.Text.Split(';');
string[] exps2 = expo2Tbx.Text.Split(';');
StreamingLearning sl = new StreamingLearning();
foreach (string keyword in keywords)
foreach (string maxOc in maxOccurences)
foreach (string minOc in minOccurences)
foreach (string exp1 in exps1)
foreach (string exp2 in exps2)
{
string learning = keyword + "_" + minOc + "_" + maxOc + "_" + exp1 + "_" + exp2;
LearningParameters learningParams = new LearningParameters(learning);
sl.TrainPredictAndWrite(trainFilePath, testFilePath, testFilePath + "pred_1.1_" + learning, sample, learningParams);
}
}
