neesetifa / bert_classification Goto Github PK

报错代码片段：
train_l_final = []
label_l_final = []
for k in train_x.keys():
train_l_final.append(train_x[k])
label_l_final.append(train.loc[k]['label'])
df_train = pd.DataFrame({'emb': train_l_final, 'label': label_l_final, })
df_train.head()

报错信息：

AttributeError Traceback (most recent call last)
Cell In[53], line 5
3 for k in train_x.keys():
4 train_l_final.append(train_x[k])
----> 5 label_l_final.append(train.loc[k]['label'])
7 df_train = pd.DataFrame({'emb': train_l_final, 'label': label_l_final, })
8 df_train.head()

AttributeError: 'function' object has no attribute 'loc'

HIERARCHICAL TRANSFORMERS FOR LONG DOCUMENT CLASSIFICATION 这篇文章除了frequency model，average model，lstm model ，还有后面一个transformer model(TOBERT)。

请问这个如何实现，能否补充下，感谢！

大佬，基于你blog中的中英文测试集。
我分别做了实验，发现用lstm效果挺差了：

英文：
2020-08-25 10:46:41,055 - long_text_classification.py[line:874] - INFO: ====== train: bert finetune Accuracy: 0.945222, Loss: 0.212253
2020-08-25 10:46:41,055 - long_text_classification.py[line:875] - INFO: ====== val: bert finetune Accuracy: 0.944176, Loss: 0.220052

2020-08-25 10:48:51,343 - long_text_classification.py[line:791] - INFO: Val Loss: 1.626610, val_acc:0.880488
2020-08-25 10:48:51,565 - long_text_classification.py[line:850] - INFO: test: lstm final Accuracy: 0.899270, Loss: 1.617715

我用我自己的一个中文测试集：

2020-08-21 20:00:32,632 - long_text_classification_zh.py[line:929] - INFO: ====== eval: bert finetune Accuracy: 0.119825, Loss: 0.856958
2020-08-21 20:00:32,633 - long_text_classification_zh.py[line:930] - INFO: ====== test: bert finetune Accuracy: 0.117358, Loss: 0.883088

2020-08-21 20:08:05,607 - long_text_classification_zh.py[line:909] - INFO: val: lstm final Accuracy: 0.838039, Loss: 1.071828
2020-08-21 20:08:05,608 - long_text_classification_zh.py[line:910] - INFO: test: lstm final Accuracy: 0.844314, Loss: 1.065375

求大佬指点