PasML

A collection of machine learning algorithms for object pascal.

Implemented algorithms

Category	Type	Model	Notes
Unsupervised	Clustering	K-Means	Only random centroid initialization. Gladly accept your contribution.
Supervised	Classification	Decision Tree	C4.5 implementation. Continuous data handling only for now.
Supervised	Classification	Naive Bayes	Gaussian naive bayes
Supervised	Regression	MLP Regressor	Multi-layer perceptron regressor. Requires noe.

Numerik compatibility

Dataset := ReadCSV('datasets/iris.csv');
X := Dataset[[_ALL_, [0, 1, 2, 3]]];
y := Dataset[[_ALL_, 4]];

Clustering example

kmeans := TKMeans.Create(3);
kmeans.Fit(X);
WriteLn('Clustering result:');
PrintMultiArray(kmeans.Predict(X));

Classification example

nb := TNaiveBayesClassifier.Create;
nb.Fit(X, y);
pred := nb.Predict(X);

WriteLn('Accuracy:');
WriteLn(Mean(pred = Ravel(y)).Item);

Integration with gnuplot

PasML has gnuplot integration support. Please ensure that you have installed gnuplot and add the executable path to the environment variable. Currently only scatter and line plots are supported.

fig := TFigure.Create('Iris Flower Features', 'Petal length', 'Petal width');
fig.AddScatterPlot(X[[_ALL_, [2, 3]]].SliceBool([y = 0]), 'Iris Setosa');
fig.AddScatterPlot(X[[_ALL_, [2, 3]]].SliceBool([y = 1]), 'Iris Virginica');
fig.AddScatterPlot(X[[_ALL_, [2, 3]]].SliceBool([y = 2]), 'Iris Versicolor');
fig.LegendPosition := lpLeftTop;
fig.Show;

fig.Free;

Note

PasML requires numerik, so you should install it first. Refer to numerik installation guide.
If you want to work with neural network, please check noe framework that was designed specifically for this task.