Time Series Prediction with Multilayer Perceptron

This is a part of Assignment 1 for Course DD2437 - Artificial Neural Networks and Deep Architectures at KTH.

1. Setup

This codebase use poetry as the python environment management tool, please have it properly installed (with Python 3.8) before continuing.

2. Data

Our dataset is a Mackey Glass series

3. Running experiments

There are 3 set of experiments:

• Grid search to find an optimal three-layer perceptron architecture
• Greedy search to fina an optimal network with best performance on noisy data
• Greedy search to find a good regularization lambda under the present of additive Gaussian noise in the training data

More details could be found in experiment.py