syxu828 / crosslingula-kg-matching Goto Github PK

您好，关于嵌入向量的获取，paper 里说的是：先由 fastText 得到嵌入向量，然后再转换到同一嵌入向量空间中。但我在 preprocessor.py 里看到使用的嵌入向量值是从文件 sub.glove.300d 中获取的，请问这个文件里的嵌入向量是将其它语言经过谷歌翻译后由模型 Glove 得到的吗？但这就与 paper 所说有所出入，麻烦您解答一下，谢谢！

A space is missing here.

As mentioned in the tile, does the transpose do something inside the tensor? Or is it just a joke or typo?

Hi, thx for sharing the code! It seems that the preprocessor.py file is missing?

Many thanks,
Weixin.

如题，谢谢啦

Hi, I just run a model 'zh_en' following the README file. However, the results on the test data are as follows:
acc @1 on Test:0.5260952380952381
acc @10 on Test:0.7301904761904762

I confuse that whether the acc@k is the same as hit@k?
If it is the same, the results ran on my machine are much worse than you reported in paper. Could you give me some tips to reproduce the results?

Thank you for your notice.

I downloaded the DBP15k.zip and found id_features file under each task directory.

For example:

zh_en\id_features_1

0	Abdullah Gul
1	Kim Dae Jung
2	National Front France

zh_en\id_features_1

10500	Abdullah Gül
10501	Kim Dae jung
10502	National Front France

It seems you initialize non-English KG's node embeddings (zh, ja, fr) with the English name feature.

How do you get the English node name for non-English KGs?

i do not find the code named preprocessor.py. please help me

The README says:

You may also choose to directly download these files from https://drive.google.com/open?id=1dYJtj1_J4nYJdrDY95ucGLCuZXDXI7PL and directly use them to train the model.

While this worked previously, recently the URL is no longer accessible. Any chance to fix this? We are using the data in PyG (pyg-team/pytorch_geometric#3980) and would love to continue supporting this dataset :)

我检查了preprocess中的代码，发现实体特征并非由论文中所描述的那样, 由跨语言词向量生成
而是通过id_features_1和id_features_2文件中的英文，查找同一英文词向量生成
据我观察id_features_1 和 id_features_2是通过机器翻译手段得到的，而论文中完全没有相关表述
这个代码实现与论文中差异过大，我是否在哪里理解错误？

您好，想请问一下是否有获取该文件的其他方式？按照提供的途径尝试多次，都因连接超时下载失败，是否有国内的链接或者能否以邮件方式获取？[email protected]，不胜感激！

Hi we are a group at LinkedIn and would like to repeat this work. This file is missing. Could you help us find the correct file ?

我怎么样才能获得preprocessor.py

When I run this code on tensorflow with big knowledge graph I get this exception:

Traceback (most recent call last):
File "/content/gdrive/My Drive/Crosslingula-KG-Matching/run_model.py", line 405, in
main()
File "/content/gdrive/My Drive/Crosslingula-KG-Matching/run_model.py", line 131, in main
train_loss = train_step(g1_v_batch, g2_v_batch, label_v_batch, if_pred_on_dev=False)
File "/content/gdrive/My Drive/Crosslingula-KG-Matching/run_model.py", line 99, in train_step
watch, _, loss = model.act(sess, "train", dict, if_pred_on_dev)
File "/content/gdrive/My Drive/Crosslingula-KG-Matching/model.py", line 620, in act
results = sess.run(output_feeds, feed_dict)
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py", line 958, in run
run_metadata_ptr)
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py", line 1181, in _run
feed_dict_tensor, options, run_metadata)
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py", line 1359, in _do_run
run_metadata)
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py", line 1384, in _do_call
raise type(e)(node_def, op, message)
tensorflow.python.framework.errors_impl.ResourceExhaustedError: 2 root error(s) found.
(0) Resource exhausted: OOM when allocating tensor with shape[32,7364,1105,400] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc
[[node node_level_matching/match_graph_1_with_graph_2/Mul (defined at content/gdrive/My Drive/Crosslingula-KG-Matching/graph_match_utils.py:10) ]]
Hint: If you want to see a list of allocated tensors when OOM happens, add report_tensor_allocations_upon_oom to RunOptions for current allocation info.

 [[aggregation_layer/second_gcn/embedding_lookup/Identity/_839]]

Hint: If you want to see a list of allocated tensors when OOM happens, add report_tensor_allocations_upon_oom to RunOptions for current allocation info.

(1) Resource exhausted: OOM when allocating tensor with shape[32,7364,1105,400] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc
[[node node_level_matching/match_graph_1_with_graph_2/Mul (defined at content/gdrive/My Drive/Crosslingula-KG-Matching/graph_match_utils.py:10) ]]
Hint: If you want to see a list of allocated tensors when OOM happens, add report_tensor_allocations_upon_oom to RunOptions for current allocation info.

您好，我在运行您的代码时发现无法重现论文的结果。我使用的参数和您在configure.py中的一样，运行的任务是zh-en。我在看代码的时候发现sample_size_per_layer设置为1，我想知道是不是因为这个才导致无法和论文描述的结果一致？这个参数值会导致每个主题图只包含一个邻结点，无法充分利用一阶领域的信息。请问您在实验时是如何设置这个参数的？还是说是其他的参数需要修改，才能达到论文的效果？