Hey can you make a video on unsupervised temporal action localization in video. like when you search google for somthing and they show time intervals where video content matches your query. I think its a great topic and may spark your interest. BTW great content as usual
Hello , i'm trying to reproduce your exercice. But i got a problem when i tried to import BERTOPIC " from import bertopic ".I get this error " no module named "llvmlite.binding.dylib". And i could not fix it; Si i wonder if you have a solution ?
Sure thing. The goal is to make a series in BERT training and code soon from scratch. In the mean time, maybe you’ll enjoy the playlist called “Transformers from Scratch” where I build a translator for a non-English language. Though there is no “pre-training” and “fine tuning”, many components of BERT are similar to the transformer. So I recommend checking that out
My spidey-sense tingles for me when more than half of the topics in the corpus are unclustered. Exploratory data analysis on those might reveal some easily fixed errors. Like, I might want to see what happens when you topic model just those bad bois, letting all the rest of the good bois to go home on the regular schoolbus. Maybe the Breakfast Club misfits have something in common with each other after all.
Have you tried to work with BERTopic with datasets that are bigger in size ? For instance 100k, 500k data sizes ? From what I have seen, the sentence transformer takes a lot of time to create the N dimensional embeddings . I am not sure if berTopic runs things in parallel.
Good question. Sorry I'm late to this. I've worked with Sentence Transformers in general and i can say the speed and quality really depends on which sentence transformer you use. You'll just need to choose the one that balances both qualities if speed is Essential (like for online applications as opposed to postmortem analysis) Also BERT and hence the sentence transformer process information in parallel. I know Sentence Transformers have this method called "embedding()" where you can pass in a list and we fetch the embeddings in parallel
Couldn't find a single article explaining the codes behind bertopic, the explanation in this video is absolutely perfect thanks!!
You're welcome :)
Another clear explanation of multiple sota useful concepts, cheers man I really like your videos and way of communicating things !
I just wanted to say that I love your videos!
This is amazing, thank you, you hero
Nice video..
Can you please explain how sentence transformer works at inference time when we have only one sentence?
Great video!!!🤗🤗🤗
Thanks a ton :)
Awesome job, man! Absolutely horrific! Thank you!
Amazing content bro. Please keep updating your playlists.
Brother will you explaint bart for text summarization
love ur bert/nlp contents
Glad you do. I'll try making more of this :)
Hey can you make a video on unsupervised temporal action localization in video. like when you search google for somthing and they show time intervals where video content matches your query.
I think its a great topic and may spark your interest. BTW great content as usual
Thanks for making this video! Helps a lot
Amazing video bro. Very helpful. Thanks so much
You are very welcome :)
Very clear explanation. Thank you :)
can u make a video about Google-Palm?
Really well explained
Thanks :)
Hello , i'm trying to reproduce your exercice. But i got a problem when i tried to import BERTOPIC " from import bertopic ".I get this error " no module named "llvmlite.binding.dylib". And i could not fix it; Si i wonder if you have a solution ?
Just a doubt. So tripplet dataset is better for improving embeddings. May a video of how to fine tuning a non english transformer?
Sure thing. The goal is to make a series in BERT training and code soon from scratch. In the mean time, maybe you’ll enjoy the playlist called “Transformers from Scratch” where I build a translator for a non-English language. Though there is no “pre-training” and “fine tuning”, many components of BERT are similar to the transformer. So I recommend checking that out
@@CodeEmporium thank you to take your time in resondijg my message. Maybe a link suggestion my friend.
Transformers from scratch
th-cam.com/play/PLTl9hO2Oobd97qfWC40gOSU8C0iu0m2l4.html
Hi, How can use BERT to word embedding
My spidey-sense tingles for me when more than half of the topics in the corpus are unclustered. Exploratory data analysis on those might reveal some easily fixed errors. Like, I might want to see what happens when you topic model just those bad bois, letting all the rest of the good bois to go home on the regular schoolbus. Maybe the Breakfast Club misfits have something in common with each other after all.
Aye good eye. I was mostly just trying to illustrate BERT. But some analysis on this would have been nice too as a follow up
Great video
Thank you 😊
Have you tried to work with BERTopic with datasets that are bigger in size ? For instance 100k, 500k data sizes ? From what I have seen, the sentence transformer takes a lot of time to create the N dimensional embeddings . I am not sure if berTopic runs things in parallel.
Good question. Sorry I'm late to this. I've worked with Sentence Transformers in general and i can say the speed and quality really depends on which sentence transformer you use. You'll just need to choose the one that balances both qualities if speed is Essential (like for online applications as opposed to postmortem analysis)
Also BERT and hence the sentence transformer process information in parallel. I know Sentence Transformers have this method called "embedding()" where you can pass in a list and we fetch the embeddings in parallel
You should be able to do this with the CLS token embeddings, instead of sentence embeddings from S-BERT, if you use regular BERT right?
in love ** divine * :D