def val_computation(alpha, beta, kappa):
# Compute the matrix and val n_train time
nmi_list, pk_list, pk_rdm_list, wd_list, wd_rdm_list = [], [], [], [], []
for _ in range(n_tests):
# Compute the membership matrix
res_matrix = spatial_clustering(d_ext_mat=d_ext_mat,
exch_mat=exch_mat,
w_mat=w_mat,
n_groups=n_groups,
alpha=alpha,
beta=beta,
kappa=kappa,
known_labels=known_labels)
# Compute the groups
alg_group_vec = np.argmax(res_matrix, 1) + 1
rstr_alg_group_vec = np.delete(alg_group_vec, indices_for_known_label)
# Compute nmi score
nmi = normalized_mutual_info_score(rstr_real_group_vec, rstr_alg_group_vec)
nmi_list.append(nmi)
# Segmentation evaluation
pk, wd, pk_rdm, wd_rdm = seg_eval(alg_group_vec, real_group_vec)
pk_list.append(pk)
pk_rdm_list.append(pk_rdm)
wd_list.append(wd)
wd_rdm_list.append(wd_rdm)
return np.mean(nmi_list), np.mean(pk_list), np.mean(pk_rdm_list), np.mean(wd_list), \
np.mean(wd_rdm_list)
# Loop on chunk
algo_group_vec = []
for chunk_id, token_chunk_list in enumerate(token_list_list):
# words x documents probabilities
word_likelihood = (norm_word_array *
np.outer(norm_document_array[chunk_id, :],
np.ones(norm_word_array.shape[1]))).T
word_groups = np.argmax(word_likelihood, 1) + 1
# Contruct the algo_group_vec
algo_chunk_group_vec = []
actual_g = 1
for w in token_chunk_list:
if len(np.where(np.array(model_voc) == w)[0]) > 0:
actual_g = word_groups[np.where(
np.array(model_voc) == w)[0][0]]
algo_chunk_group_vec.append(actual_g)
algo_group_vec.extend(algo_chunk_group_vec)
# NMI
nmi = normalized_mutual_info_score(real_group_vec, algo_group_vec)
# Segmentation evaluation
pk, wd, pk_rdm, wd_rdm = seg_eval(algo_group_vec, real_group_vec)
# Writing results
with open(results_file_name, "a") as output_file:
output_file.write(f"{input_text_file},{n_groups},{chunk_size},{nmi},"
f"{pk},{pk_rdm},{wd},{wd_rdm}\n")