Develop a sentiment classifier using feed-forward neural networks for the Twitter sentiment analysis dataset. We expect you to experiment and develop more than one models (at least two) and compare them over F1 score, Recall and Precision. For the development of the models, experiment with:
- the number of hidden layers, and the number of their units
- the activation functions (only the ones presented in the lectures)
- the loss function
- the optimizer, etc
We encourage you to use fine-tuning techniques. In the end, choose your best model and describe the reasons why you concluded to that architecture, based on the theory you were taught. For the best model plot the loss vs epochs and the ROC curve and explain what you see. Also, compare it with the solution of your logistic regression from Homework 1.