Group assignment 2

Portfolio Exercise 2:

Introduction

Task

• Build, train, and evaluate 2 special types of networks neural network with Pytorch.
  1. A CNN for a spatial prediction problem
  2. A RNN or LSTM for a sequential problem
• Experiment with at least 2 different variations of hyperparameters for each network.
• Optional: Use gradio to build a simple interactive demo (in the notebook).

This should include:

1. Feature selection and/or engineering (if necessary)
2. SPreprocessing (if necessary)
3. Train-test split.
4. Defining a neural network architecture in putorch
5. Define a training loop.
6. training the model.
7. Try out different hyperparameters.
8. Evaluate the final model on the test data.
9. Visualize results with Grad.io

Data

• 2 self choosen datasets not used in M4 so far.
• One dataset on image or other spatial data, eg. the classical MNIST dataset
• One dataset on timeseries from last semesters M1 timeseries forecast part, or M2 text classification.

Delivery

• Create a github repository (or use the existing one and adapt it)
• Save colab notebook in the github.
• Provide a readme.md with brief description.
• Submission can be in groups up to 3.
• Submit by sending an email with link to repo to Hamid (hamidb@business.aau.dk) with Daniel & Roman in cc. (dsh@…, roman@…)