def test_kl2(self):
x_probs = np.array([0.04, 0.16] * 5 + [0])
xy_probs = np.array([0.02, 0.18] * 5 + [0])
dkl = 0
for i in range(len(x_probs) - 1):
div = x_probs[i] / xy_probs[i]
dkl += x_probs[i] * math.log(div, 2)
self.assertAlmostEqual(entropy.kullback_leiber_divergence(x_probs, xy_probs), dkl)
def main(features_fpath, classes_fpath, use):
classes = np.loadtxt(classes_fpath)
if use in {'user', 'tags', 'cat'}:
prob_class, prob_col, prob_class_col = load_text_file(features_fpath,
classes, use)
else:
prob_class, prob_col, prob_class_col = load_svm_file(features_fpath,
classes)
info_gains = []
mutual_info = 0
for token in prob_class_col:
dkl = kullback_leiber_divergence(prob_class_col[token], prob_class)
mutual_info += prob_col[token] * dkl
info_gains.append((dkl, token))
print('Mutual info: ', mutual_info)
for dkl, token in sorted(info_gains, reverse=True):
print(dkl, token)
def main(features_fpath, classes_fpath, use):
classes = np.loadtxt(classes_fpath)
if use in {'user', 'tags', 'cat'}:
prob_class, prob_col, prob_class_col = load_text_file(
features_fpath, classes, use)
else:
prob_class, prob_col, prob_class_col = load_svm_file(
features_fpath, classes)
info_gains = []
mutual_info = 0
for token in prob_class_col:
dkl = kullback_leiber_divergence(prob_class_col[token], prob_class)
mutual_info += prob_col[token] * dkl
info_gains.append((dkl, token))
print('Mutual info: ', mutual_info)
for dkl, token in sorted(info_gains, reverse=True):
print(dkl, token)
def test_kl3(self):
x_probs = np.array([0.25, 0.20, 0, 0.55])
xy_probs = np.array([0.20, 0, 0.25, 0.55])
self.assertAlmostEqual(entropy.kullback_leiber_divergence(x_probs, xy_probs), float("inf"))
