Day 1 - Predicting Sequences with Neural Nets
Practical info
Place: DHØ 1.23 Time: 8:15 (we start 15 min later) - 13:20
We will be looking at LSTMs and financial forecasting. There will be also space for you to pitch assignment ideas and receive feedback.
Schedule for the day
| Time | Activity |
|---|---|
| 8:15-9:15 | Recap ANN |
| 9:30-11:00 | LSTM stock prediction |
| 11:10-12:00 | Assignment Idea Pitch |
| 12:00-12:30 | Lunch |
| 12:30-13:20 | LSTM stock prediction |
Context and Data
We will start the day with a quick recap of the AirBnb assignment. After that we will look at how LSTMs can be used to predict sequences (financial data). We will pull data from Yahoo Finance using data-libraries rather than hosted files. If you want to work with “more professional” data, you will have to get it on your own.
1. Build a baseline LSTM
Build an LSTM net that predicts closing price changes 1 day ahead
2. Extend to multi-step
Build an LSTM net that looks n timesteps back to predict the next period
3. Extend to multi-step and multi-feature
Build an LSTM that uses several inputs (e.g. other stocks or TA features)