Get sentence representation from customized dataset

[ludens11]

Hello, currently im trying to reproduce your model with different language dataset for my learning purpose. My dataset only contain lot of sentences seperated with a new line. There is no target sentences in my dataset.

the question is, with my own dataset :

• is it possible to train sent2vec like what you did in this project? because as far as i understand, you build the sent2vec model with source and target sentence. By that, it feel like my data doesnt meet the requirement.
• in Sent2Vec model class, you loaded previous rnn weight. By that, i need to use your pretrained model. Is there a way to reproduce the weight with my own dataset?

Thanks in advance

[wasiahmad]

You closed the issue, did you find answers to your questions?

[ludens11]

not yet, but im pretty sure that my data wont fit in this implementation. maybe?

[wasiahmad]

Yes, you are right. We train sentence encoder based on supervised learning and this requirement doesn't meet by your dataset. I think you should consider GenSen, USE, LASER.