How to get(save) the tensor of the entity (or hr) with the trained SimKGC model?

[EurusNotes]

您好，Liang. 请问如何用已经训练好的SimKGC模型获得 entity（或者head+rel）的tensor并保存？ 我想要向模型输入字符串然后获得它的tensor，并保存这一对pair。类似于{entity:tensor}。

• 我看了您的 evaluate.py ，请问是不是用predictor.predict_by_examples（）来获取tensor？ https://github.com/intfloat/SimKGC/blob/2139ac1573970bd93a4ce261f815da6b31f62b6a/evaluate.py#L133-L143

不过这里的 predictor.predict_by_entities 好像也可以获得tensor，对吗？我有点混乱，不知道改用哪一个好。 https://github.com/intfloat/SimKGC/blob/2139ac1573970bd93a4ce261f815da6b31f62b6a/evaluate.py#L108-L125

• 以及，您的代码是否仅适用于在终端中运行？因为我尝试用 Jupyter Notebook 的时候会发现很多地方无法运行。

希望您百忙之中能够回复我，这将能够非常帮助我。

---

Hi, Liang

How can I get the tensor data of the entity (or hr) with the SimKGC model already trained?

I read your code. Is it possible to use predictor.predict_by_examples() to get the tensor?

And, Is your code only suitable for running in the terminal this time? Because when I try to use Jupyter Notebook I find that it doesn't work in many places.

I hope you can reply to me despite your busy schedule, it would be a great help to me.

[intfloat]

Hi,

The entity_tensor corresponds to the vector representations of all the candidate entities. The hr_tensor corresponds to the vector representation of head entity + relation.

The reason there are two similar functions predict_by_examples and predict_by_entities is that predict_by_entities only computes vector for candidate entities, and can thus speed up the evaluation a bit.

I never tried to run this code base in Jupyter Notebook, but I suspect it should work fine if you import the required functions and libraries.

Thanks.

[EurusNotes]

Hi， Liang. 感谢您的回答，您的回答对我非常有帮助，让我成功地计算了 entity 的 tensor。非常感谢您。

我注意到，在预测 head entity 的时候，应该使用 tail entity + ‘inverse’ + relation 的 tensor。请问您如何获得 tail entity + ‘inverse’ + relation 的 tensor？是否也需要使用 hr_tensor？

此外，请问您：candidate entities 的数量是否等于 集合 entity 数量 - 1？

---

Thank you for your response. Your answer has been very helpful to me and I was able to successfully calculate the tensor of the entity. Thank you very much.

I noticed that when predicting the head entity, the tensor of tail entity + 'inverse' + relation should be used. Could you please tell me how to obtain the tensor of tail entity + 'inverse' + relation? Is thehr_tensor also needed?

In addition, could you please confirm if the number of candidate entities is equal to the number of entities in the set minus one?

[EurusNotes]

Hello, Liang I solved the above problems. Thank you very much for your ongoing reply. 🙏 I wish you good health.