Not find SWEM-hier

[chenghuige]

Not find SWEM-hier

[dinghanshen]

Sure, I will merge the hierarchical pooling encoder into the model.py file soon.

[cuteapi]

+1 Very interested to see it :)

[ariwaranosai]

is there any progress on this issue?

[hanhao0125]

any progress？thanks @dinghanshen

[OliverKehl]

still looking forward to this. thanks @dinghanshen

[pemywei]

still looking forward to this. thanks @dinghanshen

[ericxsun]

Still looking forward to this. thanks @dinghanshen

[qichaotang]

Still looking forward to this. thanks @dinghanshen

[LittleSummer114]

Still looking forward to this. thanks @dinghanshen

[hanxiao]

please refer to the level-mean-max for hierarchical pooling: https://github.com/hanxiao/tf-nlp-blocks/blob/8f14a864a66f976857adc04a5f3f0797dd877731/nlp/pool_blocks.py#L26

It's part of a bigger project called tf-nlp-block

[windpls]

Still looking forward to this. thanks @dinghanshen
Or, could you tell what's the stride when setting local window size = 5?

[beyondguo]

read through the paper, I didn’t find what w2v embedding other models(such as LSTM,CNN) are using. It is amazing that SWEM -ave can achieve better results than LSTM or CNN in some tasks, which in fact I don’t believe! I have done a lot of nlp tasks and I know that simply average the word embedding of a text is usually very poor.
I don’t think the comparisons of other models are fair. They don’t even use the same pretrained w2v. So maybe it’s just the Glove you used is better than the embedding other models used.

[LittleSummer114]

Hi The author gave me the swer-hier embedding, but I do not re-run it, I am also confused why so simple operation can achieve so good performance. However, our group recently finish some experiments, actually simple-operation can achieve comparable performance. For me, if you don not believe this result, you can forget this paper or you can re-run it to show whether you are right or not. Best Regards,

…

---Original--- From: "beyondguo"<[email protected]> Date: Wed, Jul 3, 2019 00:47 AM To: "dinghanshen/SWEM"<[email protected]>; Cc: "LittleSummer114"<[email protected]>;"Comment"<[email protected]>; Subject: Re: [dinghanshen/SWEM] Not find SWEM-hier (#2) read through the paper, I didn’t find what w2v embedding other models(such as LSTM,CNN) are using. It is amazing that SWEM -ave can achieve better results than LSTM or CNN in some tasks, which in fact I don’t believe! I have done a lot of nlp tasks and I know that simply average the word embedding of a text is usually very poor. I don’t think the comparisons of other models are fair. They don’t even use the same pretrained w2v. So maybe it’s just the Glove you used is better than the embedding other models used. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[JayYip]

Hi The author gave me the swer-hier embedding, but I do not re-run it, I am also confused why so simple operation can achieve so good performance. However, our group recently finish some experiments, actually simple-operation can achieve comparable performance. For me, if you don not believe this result, you can forget this paper or you can re-run it to show whether you are right or not. Best Regards,
…
---Original--- From: "beyondguo"[email protected] Date: Wed, Jul 3, 2019 00:47 AM To: "dinghanshen/SWEM"[email protected]; Cc: "LittleSummer114"[email protected];"Comment"[email protected]; Subject: Re: [dinghanshen/SWEM] Not find SWEM-hier (#2) read through the paper, I didn’t find what w2v embedding other models(such as LSTM,CNN) are using. It is amazing that SWEM -ave can achieve better results than LSTM or CNN in some tasks, which in fact I don’t believe! I have done a lot of nlp tasks and I know that simply average the word embedding of a text is usually very poor. I don’t think the comparisons of other models are fair. They don’t even use the same pretrained w2v. So maybe it’s just the Glove you used is better than the embedding other models used. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

Hi, could you share the code to me? Thanks.

[LLIKKE]

Still looking forward to this. thanks

[LittleSummer114]

感谢您的来信。