Read Me

Description

This project is a Python implementation of a variety of network types found in the paper _"Machine learning methods for postprocessing ensemble forecasts of wind gusts: A systematic comparison" by B. Schulz and S. Lerch. The use case for these networks is postprocessing of ensemble forecasts or generating probabilistic forecasts from ensemble data.

Structure of the project

There are two main directories:

1. the source code src
2. the results results

The src directory consists of both standalone Python modules, helper modules used in multiple other modules and dummy test modules of no particular interest.

src

Helpler modules

1. preprocessing.py - this module contains many functions helpful for converting the "raw" .csv files into pandas DataFrames of the correct format, scaling variables, categorizing them, etc. The exact functionality is best understood by looking into standalone Python modules such as bqn.py and looking at the data structure of the DataFrames
2. tuner_helper.py - this module contains some plotting functionality, useful to analyze the multiple runs of a model type, and is used in the tuner_analysis modules

Standalone modules

1. bqn.py, bqn_tuner.py, bqn_tuner_analysis.py - implementation of the Bernstein Quantile Network, its hyperparameter tuning and the analysis of it
2. data_exploration.py - a small script to generate a plot of the wind_power data, used in the HackMD
3. drn.py - a first implementation of the Distributional Regression Network, not furthered into a big analysis because it seamed to be too inflexible, as can be seen in the forecast plots it generates
4. emos.py, EmosModel.py - emos.py contains the first and failed attempt to work with the pywatts framework
5. hen.py, hen_tuner.py, hen_tuner_analysis.py - implementation of the Histogram Estimation Network, its hyperparameter tuning and the analysis of it
6. scoringRules.R - an import file, needed for EMOS

Dummy modules

These files only helped the author to gain an understanding of what he was doing.

• crps_test.py
• HyperParameterTest.py, HyperParameterTest2.py
• mean_model.py
• swedenExploration.ipynb, windDataExploration.ipynb
• xarray_tests

results

This folder is pretty self-explainatory