Artificial Neural Networks
4th Assignment - Shahid Beheshti University - Master’s Program April 25, 2023
Due date: May 12
- What are the main difficulties when training RNNs? How can you handle them?
- How can 1D convolutional layers be beneficial when used in conjunction with RNNs?
- What are the pros and cons of using a stateful RNN versus a stateless RNN?
- Why do people use encoder–decoder RNNs rather than plain sequence-to-sequence RNNs for automatic translation?
- Implement a time series prediction model using RNNs to forecast the stock prices of Tesla company. You will use historical data of the stock prices for the 5 years from 2016 to 2021 which can be downloaded from here. Consider the following points when tackling the problem:
- Use an appropriate train-test split strategy for time series data.
- Preprocess the data to make it suitable for RNNs.
- Experiment with different RNN variants (e.g. LSTM, GRU) and hyperparameters to achieve the best performance.
- Implement both univariate and multivariate models and compare their results.
- Evaluate the performance of the model on the test dataset. Use appropriate metrics such as MAE, MSE, or R2 score.
- Visualize the predicted vs. actual stock prices to understand the model’s performance.
- Compare the performance of the RNN model with traditional time series prediction models such as ARIMA and Prophet. (Extra point)