Assignment 2

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 2

Applied AI and Control

IP505245 Applied AI and Control


Assignment 2, A trajectory prediction model
Out: 05.10.2022
In: 22.10.2022

1) Use the provided AIS data to train a seq2seq model and evaluate the performance.
• The input for this model should be 24 time steps (4 min) and the output for this model
should be 12 time steps (2 min).
• The encoder and decoder of the seq2seq model should have 2 hidden layers, and each
hidden layers should have 64 hidden units.
• Plot the training curves (training epoch vs. MSE loss) and show the average distance error
and final distance error from the test set.

2) Optimize the model (investigate the effects of different parameters). Note that for fair comparison,
do not change the output time steps, which should be 12 time steps (2 min).
• Adjust the learning rate (1e-2, 1e-3, 1e-5) and plot the training curves (training epoch vs.
MSE loss) and show the ADE and FDE.
• Keep a constant learning rate, adjust the network architecture (n_layer, n_hidden_unit) and
show the performance of different configurations.
• Optional*: Try different input length (maximum 100 time steps) and show the performance
of the model when using different input length.

Hint:

1. The Python code / Jupyter notebook code in lecture 7 can be used as the reference for this
assignment.
2. Note you need a library called pyproj in the utlis,py in the code template for Assignment 2 to
process the data. This library is used to transform the latitude and longitude of the vessel to (x, y) in
meters.
3. You can directly call the preprocess() function in utlis.py in the code template for Assignment 2 to
preprocess the raw data.

1 of 2
Applied AI and Control
4. You can directly use the metrics() function in utlis.py in the code template for Assignment 2 to
evaluate the model performance from test data.
5. The final report should consist of the following parts:
a. A short description of the dataset
b. A short description of how the dataset is preprocessed and split into train and test set
c. How the model is built and the performance of the model (Q1 above).
d. How different hyperparameters affect the performance of the model (Q2 above).

2 of 2

You might also like