Objeck Documentation

Bundle

Core machine learning types: the seedable Random generator, Matrix2D linear algebra and activations, matrix reference holders, the NeuralNetwork and CSV-backed MatrixReader. Compile with -lib ml.

TableEncoder

Turns a CsvTable into the numeric matrix a model can train on, learning its encoding once so the same encoding can be reapplied to other tables. MatrixReader already loads a CSV, but only one that is numeric throughout, and it reads the header row as another record whose non-numeric cells become 0.0. Real tabular data is mixed: NSL-KDD's protocol, service and flag columns are strings among thirty-eight numeric ones. This skips the header, one-hot encodes the string columns and passes the numeric ones through. The vocabulary is fixed at Fit. Transform reuses it, so a training table and a test table produce the same columns in the same order even when the test split never contains some category, and a category that appears only in test encodes as all-zeros rather than silently shifting every column after it.

Example

enc := TableEncoder->New("class");
train_x := enc->FitTransform(train);
train_y := enc->TransformLabelsBinary(train, "normal");
test_x := enc->Transform(test);

Fit #

Learns the feature layout: which columns are numeric, and the value vocabulary of each column that is not. A column counts as numeric only when every data value in it parses as a number. One non-numeric value makes the whole column categorical, which is the safe direction: encoding a number as a category costs columns, while reading a category as a number invents an ordering that is not there.

method : public : Fit(table:CsvTable) ~ Bool

Parameters

Name	Type	Description
table	CsvTable	table to learn from, normally the training split only

Return

Type	Description
Bool	true when the layout was learned, false for an unparsed or empty table

FitTransform #

Fits on this table and encodes it. For the training split only.

method : public : FitTransform(table:CsvTable) ~ Float[,]

Parameters

Name	Type	Description
table	CsvTable	table to fit and encode

Return

Type	Description
Float	encoded matrix, or Nil when the table could not be fitted

GetFeatureNames #

The name of each encoded column: the source column for a numeric feature, and "column=value" for each one-hot column.

method : public : GetFeatureNames() ~ String[]

Return

Type	Description
String	names in output order, or Nil when unfitted

GetLabelClasses #

The label classes seen at Fit, in the order TransformLabels indexes them.

method : public : GetLabelClasses() ~ String[]

Return

Type	Description
String	class values, or Nil when unfitted

GetWidth #

The number of columns Transform produces.

method : public : GetWidth() ~ Int

Return

Type	Description
Int	encoded width, or 0 when unfitted

IsFitted #

Whether Fit has been called successfully.

method : public : IsFitted() ~ Bool

Return

Type	Description
Bool	true when this encoder holds a layout

New # constructor

Constructor for a table with no label column: every column is a feature.

New()

New # constructor

Constructor

New(label_col:String)

Parameters

Name	Type	Description
label_col	String	name of the label column, which is excluded from the features

Transform #

Encodes a table using the layout learned by Fit.

method : public : Transform(table:CsvTable) ~ Float[,]

Parameters

Name	Type	Description
table	CsvTable	table to encode; its header row is skipped

Return

Type	Description
Float	matrix of (data rows) x (encoded width), or Nil when unfitted or the table cannot be read

Example

test_x := enc->Transform(test);
if(test_x = Nil) {
  "unfitted, or the table did not parse"->PrintLine();
};

TransformLabels #

Encodes the label column as class indices, using the classes seen at Fit.

method : public : TransformLabels(table:CsvTable) ~ Float[]

Parameters

Name	Type	Description
table	CsvTable	table to read labels from

Return

Type	Description
Float	one index per data row; a class unseen at Fit becomes -1.0

TransformLabelsBinary #

Encodes the label column as a two-class target: 0.0 for the named value, 1.0 for everything else. Intrusion detection labels are usually one benign value against many attack names, and this avoids enumerating the attacks.

method : public : TransformLabelsBinary(table:CsvTable, negative:String) ~ Float[]

Parameters

Name	Type	Description
table	CsvTable	table to read labels from
negative	String	the label value that maps to 0.0

Return

Type	Description
Float	one value per data row, or Nil when unfitted or the label is missing