Learning to Identify Top Elo Ratings

We propose two algorithms MaxIn-Elo and MaxIn-mElo to solve the top players identification on the transitive and intransitive settings. All winning probability matrices of games are saved in file 'games/'. You can install the required packages by:

pip install -r requirements.txt

Baselines

There are 5 baselines in mrandom.py, mDBGD.py, mRGUCB.py, mELOMLE.py, MaxInELO.py.

For mrandom.py, mDBGD.py, mRGUCB.py and MaxInELO.py, we set a parameter 'self.melo' to control using Elo or mElo to update ratings.

Runs

Results of top-1 identification

For the Elo model, you can tune the best parameters of top-1 performance on transitive games by running:

sh runelo.sh 

Then you can plot the results of top-1 on the Elo model by running:

python Elo_plot.py Max 0

All figures are save in file finalplot/. And the results of top-1 of Elo showed in figure 2 are obtained:

For the mElo model, you can tune the best parameters of the top-1 performance on intransitive games by running:

sh runmelo.sh 

Then you can plot the results of top-1 on the mElo model by running:

python Elo_plot.py Max 1

The results of top-1 on mElo in Figure 3 are obtained:

Results of top-k identification

You can get the results of top-k identification of all baselines by running:

sh runelo.sh
python topk_plot.py

Comparison of different $\gamma$

You can get the results of different $\gamma$ of our MaxIn-Elo on transitive games by running:

python compare_gamma.py

Comparison of different dimension C of vectors used in mElo

You can get the results of different C of our MaxIn-mElo on an intransitive game by running:

sh run_c.sh
python C_plot.py

Comparison of different batch size $\tau$

You can get the results of different batch size $\tau$ of our MaxIn-Elo on an transitive game by running:

sh run_batch.sh
python batch_plot.py 0

Then the results of different batch size $\tau$ of our MaxIn-Elo on an intransitive game can be obtained by running:

python batch_plot.py 1