import regex

def mismatch_kernel(chainsA, chainsB, k=3, m=0):
    chainsA = np.atleast_1d(chainsA)
    chainsB = np.atleast_1d(chainsB)

    NVocab = {}
    vocab = np.array(["".join(item) for item in itertools.product("ATCG", repeat=k)])
    n_vocab = len(vocab)
    idx = dict(zip(vocab, range(len(vocab))))
    all_seq = "".join(vocab)
    for kmer in vocab:
        neighbors = regex.findall("(" + kmer + ")" + "{s<=" + str(m) + "}", all_seq, overlapped=True)
        NVocab[kmer] = list(np.unique(neighbors))

    specA = []
    for chain in chainsA:
        spec = np.zeros(n_vocab)
        n = len(chain[0])
        for offset in range(n - k):
            kmer = chain[0][offset: offset + k]
            for nb in NVocab[kmer]:
                spec[idx[kmer]] += 1
        specA.append(spec)
    specA = np.asarray(specA)

    specB = []
    for chain in chainsB:
        spec = np.zeros(n_vocab)
        n = len(chain[0])
        for offset in range(n - k):
            kmer = chain[0][offset: offset + k]
            for nb in NVocab[kmer]:
                spec[idx[kmer]] += 1
        specB.append(spec)
    specB = np.asarray(specB)

    return np.dot(specA, specB.T)