np.mean(P_val),
np.mean(R_val),
np.mean(ap_val)
]
learning_rate = lr.adapt(mAP_val)
values = [
i, l, l_rank, l_att_span, l_att_global, l_att_dist, mF1_val,
mAP_val, learning_rate
]
logger.add(values)
print(
'{} loss: {} l_rank: {} l_att_span: {} l_att_global: {} l_att_dist: {} mF1: {} mAP: {} lr: {}'
.format(*values))
print('Precision: {} Recall: {}'.format(mP_val, mR_val))
logger.save()
print('learning rate', learning_rate)
if is_save and mAP_val >= logger.get_max('mAP'):
saver.save(sess, save_path + '/model_ES.ckpt')
def evaluate_df():
ap_tst, predictions_mll, labels_mll = evaluate(
iterator_test, [img_ids_test, features_test, seen_labels_test],
model.features, model.logits, sess, model)
F1_3_tst, P_3_tst, R_3_tst = evaluate_k(
3, iterator_test, [img_ids_test, features_test, seen_labels_test],
model.features, model.logits, sess, model, predictions_mll, labels_mll)
F1_5_tst, P_5_tst, R_5_tst = evaluate_k(
5, iterator_test, [img_ids_test, features_test, seen_labels_test],
