Interpret textual data using BERT and then try to predict the direction change of the benchmark index DJIA using only BERT embeddings. Use the "bert$-$base$-$uncased" model instead of fine$-$tuning BERT. Use $\text{'}$MLP with $1$ hidden layer and $8$ units in hidden layer' to predict DJIA's direction from BERT embedding. BERT is already included in the Hugging Face library.

"Combined$\_$News$\_$DJIA.csv$"$ contains the day's news, while "DJIA$\_$table.csv$"$ contains the DJIA's adjusted close price data. We can make a sample every day. Following that, we can use an $80\%-20\%$ partition for train and test data. It is important to note that the learning rate and number of epochs must be carefully adjusted.