def neighbour_joining(self):
        D = self.arr.copy()
        n = len(D)

        if n == 2:
            return UnrootedTree((0, 1, D[0][1]))
        if n < 2:
            return UnrootedTree()

        edges = []
        labels = [i for i in range(n)]
        nextlabel = n

        while n > 2:
            # neighbour-joining matrix constructed from the distance matrix D, Q1
            sum = np.sum(D, axis=0)
            Q = np.transpose(np.transpose(D * (n - 2) - sum) - sum)
            np.fill_diagonal(Q, 0)

            # obtain row and col index of minimum element
            (rowi, coli) = np.unravel_index(Q.argmin(), Q.shape)
            if rowi > coli:
                (rowi, coli) = (coli, rowi)

            val = D[rowi, coli]

            joininglength1 = 0.5 * val + (1 / (2 * (n - 2))) * (np.sum(D[rowi]) - np.sum(D[coli]))
            edges.append((labels[rowi], nextlabel, joininglength1))         # joininglength1
            edges.append((labels[coli], nextlabel, val - joininglength1))   # val - joininglength1 = joininglength2

            # updaten van D matrix
            D[rowi, :] = (D[rowi, :] + D[coli, :] - val) / 2
            D[:, rowi] = D[rowi, :]

            labels[rowi] = nextlabel
            del labels[coli]
            D = np.delete(np.delete(D, coli, 0), coli, 1)

            nextlabel += 1
            n -= 1

        (label1, label2) = (labels[0], labels[1])
        if label1 > label2:
            (label1, label2) = (label2, label1)
        edges.append((label1, label2, D[0, 1]))

        return UnrootedTree(*edges)