Apologies for the couple of 🎤 statics in the background , but after the Min 7, it's smooth sailing. Will ensure the subsequent episodes are a great listening experience. Thanks for understanding.
@yogeshsingh9749
Жыл бұрын
Extremely valuable content. I've applied the same techniques at my workplace. Haven't seen anyone delivering this type of high quality industry grade content with hands-on approach.
@prithivida
Жыл бұрын
Thanks for the kind words. Very humbling. I will try to bring more contents in the coming months.
@prithivida
2 жыл бұрын
The link to the code / notebook is in the video description 👆🏼. For technical questions and discussions please head to the slack channel (linked in the description as well. )
@JAZEL1111
Жыл бұрын
Really liked your approach of going through simple stuff quickly and giving appropriate time to important stuff!!
@prithivida
Жыл бұрын
Thanks for the kind words.
@anujsingh5961
2 жыл бұрын
Thank you very much for all your efforts . Such a gem content 🙏
@prithivida
2 жыл бұрын
Glad you enjoy it!
@abhijeetnarharshettiwar6175
2 жыл бұрын
Hi Prithvi, You are doing amazing job. Thank you so much for such insightful video.
@prithivida
2 жыл бұрын
Thanks a ton
@anoubhav
3 ай бұрын
Why is the sentence Bert approach (on plot vs. label) wrong at 28:30? Is it due to the differing lengths (symmetrical vs. asymmetrical sentence embeddings?) Or something entirely different? thank you for your videos!
@prithivida
2 ай бұрын
Thanks for reaching out, Richer the context better the sentence embeddings, that’s why most older models in ST uses mean pooling I.e mean of the token embeddings of all the tokens in the last hidden layer. Single words like labels won’t make a lot of semantic sense. You can test it out yourself. Take Fx GoEmotions datasets where emotions are labels, try the SoTA ST models to embed text and labels, round cosine sim as preds, get F1 scores across labels and compare the same F1 scores from a fine tuned classifier.
Пікірлер: 12