Overview

This pet-project aims to compare a classical Machine Learning model, LightGBM, with Neural Networks for the time-series forecasting task in terms of dev time complexity. The task is to create several models for the day-ahead energy price forecasting using open source data for Germany, while tracking the time spent on each model with Clockify.

The project is designed to ensure all experiments are reproducible.

Structure

Deep-Learning-for-Time-Series/
├── data/
│   └── raw/
│       ├── capacities/                  # installed capacity of power generating units per energy type
│       ├── consumption/                 # energy load
│       ├── prices/                      # day-ahead energy prices
│       ├── production/                  # energy production  
│       └── weather_forecast/            # historical weather forecast
│   └── preprocessed/               
├── models/
├── notebooks/
│   ├── 01_data_exploration.ipynb        # EDA
│   ├── 02_feature_engineering.ipynb     # feature engineering & selection
├── src/
│   ├── load_data.py
│   ├── data_preprocessing.py            # scripts for data cleaning & transformation
├── requirements.txt
├── .gitignore
└── README.md