Transfer Learning: Hands On Bert 😀
Hi Medium, Lets Talk about state-of-art in Natural Language Processing. We Are going to focus on only three words →
What ? Why ? How ?
This article will only focus on Introduction and Coding part.
WHAT is BERT ?
BERT stands for Bidirectional Encoder Representations from Transformers . It is an Open-Source project by Google AI researchers with a great power of understanding the context of sentence (language) showing high performance in various nlp tasks such as question-answer system , Named-entity-recognition, Machine Translation and many more.
WHY BERT ?
Bert is based on transformer model that uses Attention mechanism for learning contextual relationship among words of a sentence i.e. it takes positional encoding into account. Lets have a view at an example below →
Sentence 1: dog bites man
Sentence 2: man bites dog
What is the difference between two ? Its the position of words ! Ohh Damn ! this is what most nlp models were missing back then . Is it all Bert have ? No !
Bert has another most important feature Masked Language Modelling and Feed Forward Layer.
Feed Forward is basically for taking using of backpropagation and introducing some non-linearity in model.
MLM — The model then attempts to predict the original value of the masked words, based on the context provided by the other, non-masked, words in the sequence
Transfer Learning → Using and modifying a pre-trained model to our needs. We are using now → https://github.com/google-research/bert
HOW ?
Github Links → https://github.com/r-sajal/DeepLearning-/tree/master/Natural-Language-Processing/Part%201
Model 1: Hugging face Transformer
You can find the comments for understanding the code. For any queries please comment
Following Three Pictures were For those who have multiclass classification instead of binary →
.
Click on the Bert folder on left Image
.
.
.
.
In this image on left open run_classifier.py
.
.
.
Put as many numbers you want in the list separated by comma representing the classes of your classification.
.
.
Model 2: Ktrain
You can find the comments for understanding the code. For any queries please comment
Reference →
UPVOTE !!!!!! Please 🙇♂️
Thank you for your Precious Time .
Personal Links →
