#for meshs in top_10_mesh:
# mesh = ' '.join(meshs)
# pred_label_10.writelines(mesh.strip()+ "\r")
#pred_label_10.close()
#pred_label_15 = open('MultichannelCLSTM_pred_label_15.txt', 'w')
#for meshs in top_15_mesh:
# mesh = ' '.join(meshs)
# pred_label_15.writelines(mesh.strip()+ "\r")
#pred_label_15.close()
end = time.time()
print("Run Time: ", end - start)
########################### Evaluation Metrics #############################
# precision @k
precision = precision_at_ks(pred, test_labelsIndex, ks=[1, 3, 5])
for k, p in zip([1, 3, 5], precision):
print('p@{}: {:.5f}'.format(k, p))
# check how many documents that have mesh terms greater and equal to 10/15
label_row_total = np.sum(test_labels, axis=1)
index_greater_10 = [
index for index, value in enumerate(label_row_total) if value >= 10
]
index_greater_15 = [
index for index, value in enumerate(label_row_total) if value >= 15
]
def get_label_using_index(org_label, index):
predicted_label = np.zeros(predictions.shape)
for i in range(len(predictions)):
top_k_index = (predictions[i].argsort()[-k:][::-1]).tolist()
for j in top_k_index:
predicted_label[i][j] = 1
predicted_label = predicted_label.astype(np.int64)
return predicted_label
top_10_pred = top_k_predicted(pred, 10)
top_20_pred = top_k_predicted(pred, 20)
end = time.time()
print("Run Time: ", end - start)
########################### Evaluation Metrics #############################
# precision @k
precision = precision_at_ks(pred, test_labelsIndex, ks=[1, 3, 5])
for k, p in zip([1, 3, 5], precision):
print('p@{}: {:.5f}'.format(k, p))
# nDCG @k
nDCG_1 = []
nDCG_3 = []
nDCG_5 = []
Hamming_loss_10 = []
Hamming_loss_20 = []
for i in range(pred.shape[0]):
ndcg1 = ndcg_score(test_labels[i], pred[i], k=1, gains="linear")
ndcg3 = ndcg_score(test_labels[i], pred[i], k=3, gains="linear")
ndcg5 = ndcg_score(test_labels[i], pred[i], k=5, gains="linear")