#if(cur_exp_param=='cpu'):
# labels_to_eval=y_test_all_labels[:,:math.floor(difference_percent*output_classes)]
#else:
# labels_to_eval=y_test_all_labels[:,:1]
if (cur_exp_param == 'cpu'):
top_k = 10
else:
top_k = 5
while i < top_k:
y_pred = test_predict_top_k[:, 0:i + 1]
i = i + 1
ndcg_i = func_eval._NDCG_score(y_pred, labels_to_eval)
ndcgs.append(ndcg_i)
print(ndcg_i)
#sys.exit(0)
print("precision:")
i = 0
precisions = []
while i < top_k:
y_pred = test_predict_top_k[:, 0:i + 1]
i = i + 1
for i, each in enumerate(test_labels):
y_real.append(np.array([each]))
y_real = np.array(y_real)
print('current exper_param is : ', dataset)
print("ncdg:")
i = 0
ndcgs = []
top_k = 5
while i < top_k:
y_pred = test_pred[:, 0:i + 1]
i = i + 1
ndcg_i = func_eval._NDCG_score(y_pred, y_real)
ndcgs.append(ndcg_i)
print(ndcg_i)
#sys.exit(0)
print("precision:")
i = 0
precisions = []
while i < top_k:
y_pred = test_pred[:, 0:i + 1]
i = i + 1
for j,label in enumerate(new_sort_list):
indices = [i for i, l in enumerate(y_test) if l == label]
each_label_real=[y_test[i] for i in indices]
# if(each_label_real==[]):
# f1_score[j]=0
# print(f1_score[j])
# continue
each_label_predict=np.array([test_predict_top_k[i] for i in indices])
each_labels_to_eval=np.array([labels_to_eval[i] for i in indices])
precision=func_eval._precision_score(each_label_predict,each_labels_to_eval)
precisions[j]=precision
recall=func_eval.new_recall(each_label_predict,each_labels_to_eval)
recalls[j]=recall
ncdg=func_eval._NDCG_score(each_label_predict,each_labels_to_eval)
ncdgs[j]=ncdg
#print('\nrecalls:',recalls)
#print('precisions:',precisions)
#print('ncdgs:',ncdgs)
#----------3. split labels into group by label frequency-------------------------
#-----------get the evaluation of each group------------------------------
group_recalls=[]
group_precisions=[]
group_ncdgs=[]
