softmaxDeriv function

Matrix softmaxDeriv(Matrix matrix)

derivative of a tanh matrix

• matrix The matrix to find the derivative
• Returns The new matrix

Implementation

Matrix softmaxDeriv(Matrix matrix) {
  Matrix softmaxValues = Matrix(matrix.row, matrix.col);
  for (int i = 0; i < matrix.row; i++) {
    List<double> row = matrix[i] as List<double>;
    double maxV = row.reduce((a, b) => a > b ? a : b);

    // Calculate softmax values for the current row
    List<double> softmaxRow = row.map((x) => exp(x - maxV)).toList();
    double sumExp = softmaxRow.reduce((a, b) => a + b);
    softmaxRow = softmaxRow.map((x) => x / sumExp).toList();

    // Fill the softmaxValues matrix
    for (int j = 0; j < matrix.col; j++) {
      softmaxValues.setAt(i, j, value: softmaxRow[j]);
    }
  }

  // Create the Jacobian matrix
  Matrix jacobian = Matrix(softmaxValues.row, softmaxValues.col);
  for (int i = 0; i < softmaxValues.row; i++) {
    for (int j = 0; j < softmaxValues.col; j++) {
      if (i == j) {
        // Diagonal elements
        jacobian.setAt(
          i,
          j,
          value: softmaxValues.getAt(i, j) * (1 - softmaxValues.getAt(i, j)),
        );
      } else {
        // Off-diagonal elements
        jacobian.setAt(
          i,
          j,
          value: -softmaxValues.getAt(i, j) * softmaxValues.getAt(j, j),
        );
      }
    }
  }
  return jacobian;
}