Numerik is a numerical library for object pascal. It supports:

• Multidimensional array data structure with single precision
• Efficient array broadcasting and multidimensional slicing
• Some arithmetic and linear algebra functionalities
• BLAS-accelerated operation

Installation

• If you use lazarus, the easiest way is to install through "numerik.source.lpk" package inside "pkg" directory. Open the package, compile, and add to project.
• Optional but strongly recommended, provide the required libraries:
  • Windows: Provide libopenblas.dll which can be obtained from here
  • Ubuntu: Install libopenblas-base libopenblas-dev liblapacke liblapacke-dev

Example

uses
  multiarray, numerik;

var
  A, B: TMultiArray;
  RowIndices, ColIndices: TLongVector;
  SVDResult: TSVDResult;

begin
  A := [1, 2, 3, 4, 5, 6]; // TMultiArray accepts open array assignment
  A := A.Reshape([3, 2]);  // Reshape it into a 3x2 array
  PrintMultiArray(A);

  { Load data from CSV }
  A := ReadCSV('data.csv'); 
  PrintMultiArray(A);
  
  { 5 x 5 random array }
  A := Random([5, 5]);
  PrintMultiArray(A);

  { Multidimensional slicing }
  RowIndices := [0, 3];
  ColIndices := [0, 1, 3, 4];
  B := A[[ RowIndices, ColIndices ]]; // Equivalent to A.Slice([RowIndices, ColIndices])
  PrintMultiArray(B);

  { Math operation }
  PrintMultiArray(Transpose(B));   // Equivalent to B.T
  PrintMultiArray(Exp(B));         // Element-wise exponentiation
  PrintMultiArray(Add(B, 2));      // Equivalent to B + 2
  PrintMultiArray(Multiply(B, B)); // Element-wise multiplication, Equivalent to B * B
  PrintMultiArray(Matmul(B.T, B)); // Matrix multiplication 
  
  { Singular value decomposition }
  SVDResult := SVD(B);
  PrintMultiArray(SVDResult.U);
  PrintMultiArray(SVDResult.Sigma);
  PrintMultiArray(SVDResult.VT);

  { Pseudo-inverse }
  PrintMultiArray(PseudoInverse(B));
end.      

Read the API reference here.

Interested in neural networks? Numerik is being integrated with Noe, neural network framework for pascal.