The hands on project on Sentiment Analysis with Deep Learning using BERT is divided into following tasks:
Task 1: An introduction to some basic theory behind BERT, and the problem we will be using it to solve
Task 2: Explore dataset distribution and some basic preprocessing
Task 3: Split dataset into training and validation using stratified approach
Task 4: Loading pretrained tokenizer to encode our text data into numerical values (tensors)
Task 5: Load in pretrained BERT with custom final layer
Task 6: Create dataloaders to facilitate batch processing
Task 7: Choose and optimizer and scheduler to control training of model
Task 8: Design performance metrics for our problem
Task 9: Create a training loop to control PyTorch finetuning of BERT using CPU or GPU acceleration
Task 10: Load in a pre-saved finetuned BERT model and evaluate its performance, and some final thoughts
The course materials belong entirely to coursera. The answers are the only things that show my trials. Special thanks to the instructor: Ari Anastassiou