def run_single_nmf_experiment(mat, n_components, apply_nan, labels):
if apply_nan:
nmf = NMF.NMF(n_components=n_components, apply_nan_mask=True, nan_weight=0.1)
else:
nmf = NMF.NMF(n_components=n_components)
W, H = nmf.decompose(mat)
clusters = W.argmax(axis=1)
nmi, nmi_t, nmi_p = calc_score_significance(labels, clusters, normalized_mutual_info_score)
if nmi_p >= 0.05:
print('non conclusive NMI with {} and nan_mask={} '.format(n_components, apply_nan),
nmi_t, nmi_p)
sill = silhouette_score(mat.toarray(), clusters)
accs = []
f1s = []
for i in range(20):
f1, f1_t, f1_p, acc, acc_t, acc_p = semi_supervised_classify(labels, clusters)
accs.append(acc)
f1s.append(f1)
if f1_p >= 0.05:
print('non conclusive f1 score with {} and nan_mask={} '.format(n_components, apply_nan),
f1_t, f1_p)
if acc_p >= 0.05:
print('non conclusive accuracy score with {} and nan_mask={} '.format(n_components, apply_nan),
acc_t, acc_p)
return {'nmi': nmi, 'nmi_p': nmi_p, 'sill': sill,
'acc': np.mean(accs), 'acc_std': np.std(accs),
'f1': np.mean(f1s), 'f1_std': np.std(f1s)}
