Fine Tuning Transformer (BERT) for Customer Review Prediction | NLP | HuggingFace | Machine Learning

[@thegreatlazydazz]

Great content. Just a suggestion, whenever you are writing a tensor for the first time, try to write the dimension of the tensor as a comment. If the dimension gets updated (for example by taking argmax) write the updated dimension on the side.

[@RohanPaul-AI]

Thanks Anirban for the suggestion, will try to include that.

[@sohambasu660]

Can this be considered a regression problem ? Since, there are reviews which are numerical in nature ?

[@orkhanamrullayev3966]

Thanks for the tutorial. For the cross-entropy requirements, we needed to use the range 0-4 instead of 1-5, could not we use just one line of code for mapping instead of a multiline function? I mean this line: df['stars'] = df['stars'] - 1
One more question, do we need to balance the size of the classes?

[@RohanPaul-AI]

Yes you could use the one liner for sure. Thats actually better.

[@rashdakhanzada8058]

How to fine tune bert for roman urdu datset?

[@compeng2013]

Could you achive this just using HuggingFace Trainer Api instead of doing it all manually via PyTorch?

[@RohanPaul-AI]

Absolutely you can with some minor code changes. I just put a file for the general guideline to do this.
Check out - github.com/rohan-paul/MachineLearning-DeepLearning-Code-for-my-RUclips-Channel/blob/master/NLP/Fine_Tuning_HuggingFace_Transformer_BERT_Yelp_Customer_Review_Predictions/Example_Code_Using_HF_Trainer_API.ipynb

[@kuberchaurasiya]

Thanks for the nice educative videos!!

[@RohanPaul-AI]

Glad you liked @kuberchaurasya

[@not_amanullah]

Zoom in can't see clearly

[@RohanPaul-AI]

Sorry about that, will resolve for sure.