To Be Implemented: Sentiment Classification with BERT embeddings
Within the classifer.py file you will find a pipeline that • Calls the BERT model to encode the sentences for their contextualized representations • Feeds in the encoded representations for the sentence classification task • Fine-tunes the Bert model on the downstream tasks (e.g. sentence classification) Within this file, you are to implement the BertSentimentClassifer. You will implement this class to encode sentences using BERT and obtain the pooled representation of each sentence8. The class will then classify the sentence by applying on dropout the pooled output and then projecting it using a linear layer. Finally (already implemented), the model must be able to adjust its parameters depending on whether we are using pre-trained weights or are fine-tuning.
see section 6.2