Match to Match Prediction

How to run the pipeline

When training over the years 2016 to 2018 for testing on the year 2019

1. Check that these files exist in the data folder

• aus-open-player-stats-2015.csv
• aus-open-player-stats-2016.csv
• aus-open-player-stats-2017.csv
• m2016.csv
• m2017.csv
• m2018.csv
• m2019.csv

2. Open the notebook neural_network.ipynb / svm.ipynb / gradient_boost.ipynb

3. Check that the Inputs heading contains the right parameters.

4. Run the notebook until a preview of the variable r_16_merge is displayed.

5. Compare it to the tournament result.

When adding more data,

• Download match outcome data from this website
• If adding data for the year 2015, Save it as m2015.csv
• Run the main.py as

python3 main.py --year 2014
python3 all_pressure.py --year 2014

• Continue from Step 2

Model used

• Neural Network
  • Hyperparameters
    • activation=logistic
    • solver=adam
    • learning_rate=adaptive
    • max_iter=10000
    • warm_start=True
    • hidden_layer_sizes=(128,50,2)
• Support Vector Machine (linear kernel)
  • Hyperparameters
    • kernel=linear
    • C=0.001
    • max_iter=1000000
• Gradient Boost
  • Hyperparameters
    • n_estimators=10000
    • learning_rate=0.00001
    • criterion=squared_error
    • loss=exponential

Assumptions

• People with no stats have been considered as walkovers

Summary

Australian Open 2019

Model	Round of 16	Round of 32	Round of 64
Baseline	8	22	43
Neural Network	10	23	50
SVM (Linear)	9	23	50
Gradient Boost	10	22	54

Australian Open 2018

Model	Round of 16	Round of 32	Round of 64
Baseline	11	20	34
Neural Network	6	17	40
SVM (Linear)	5	20	46
Gradient Boost	8	22	46