bl-predictor 1.2

Description:

blpredictor 1.2

bl-predictor










Bl-predictor is a simple application for predicting game results for the German 1. Bundesliga.
It features a clean graphical user interface (including DarkMode to spare your eyes), automatic data crawling, a variety of
prediction-models to choose from, and a few built-in model evaluation tools.

Usage
Install:
pip install bl-predictor

and start the GUI:
bl-predictor-gui

The left column shows you the next upcoming matches. These are automatically crawled from OpenligaDB when the application starts.
The center column gives you the option to tweak your prediction preferences:

choose the seasons used for training the model via the slider
select a model to train
choose a home and guest team

Your result and additional information about the model used will appear in the righthand column.
To make another prediction just use one of the reset-options on the bottom-left.

You can switch to dark mode or exit the application under "Options" in the top-left corner.

Prediction models
PoissonModel
A model that predicts the winning team out of two given teams, based on a poisson regression model.
Caution: The model is sensitive to the order of given teams, because the home_team scores better on average!
This model is heavily based on a guideline from David Sheehan.
BettingPoissonModel
A adaptation of the PoissonModel improved for betting.
If no relevant (>10%) difference in the teams winning probabilities is present, "Draw" is returned.
FrequencyModel
A model that uses all results of the last seasons to predict a winner based on the relative frequency of wins.
Model Evaluation
The model evaluation features no graphical user interface.
To access it you will need to go into the package source files to prediction_evaluation.py
and call the functions given at the bottom of the file. You can:

generate a plot about the accuracy / F1-score of all models with different trainset sizes
evaluate a single models performance

trainset information
performance measures
((Betting-)PoissonModel also returns a team-ranking based on the models coefficients)


compare two models
get general statistics about a trainset

The results will either be given as printout in the console or as plots.png and will look something like this:








License
bl-predictor is made available under the MIT-License