Homework 12: Time Series Prediction with LSTM
Homework 12
Time Series Prediction with LSTM
Check the article: Time Series Prediction with LSTM Recurrent Neural Networks in Python with Keras. It describes how to solve the International Airline Passengers prediction problem using a simple LSTM recurrent neural network. The link to download the dataset “airline-passengers.csv” is also available in this article. Download and run the code from this article. Note that you need to change “from keras…” to “from tensorflow.keras…”.
Experiment with this code trying to improve the performance. Some suggested tuning includes:
- Adding early stopping
- Adding additional LSTM layers
- Adding dropout
- Changing the number of cells per LSTM layer
- Changing the optimizer
- Changing data scaling
- The number of previous time steps (look_back) to use as input variables to predict the next time period
Report your results in a table similar to the one below:
|
No |
Experiment |
Train RMSE Score |
Test RMSE Score |
|
1 |
Base: 1 LSTM Layer, Dim=4, Adam, min-max scaling, look_back=1 |
22.83 |
50.59 |
|
2 |
|
|
|
|
3 |
|
|
|
|
4 |
|
|
|
|
5 |
|
|
|
|
6 |
|
|
|
|
7 |
|
|
|
Submit your report (in pdf format) that contains the above table of your experiments, charts, and source code of best model.