Tetris Bot Challenge

Goal

Create a Tetris bot that gets the maximum score in 10 random games.

Idea

If I create a fixed set of 1000 games and create a bot that will get the maximum average score, it will play well in 10 random games.

Methodology

1. Simulate every rotation and position​
2. Calculate objective value​
3. Select the rotation and position where the objective value is best​
4. Apply the rotation and position

Objective Value​

• Aggregate height
• Cleared lines
• Holes
• Bumpiness

Objective Function

OV = CL×CLW – AH×AHW – H×HW – B×BW

OV: Objective Value​
CL: Cleared Lines​
CLW: Cleared Lines Weight​
AH: Aggregate Height​
AHW: Aggregate Height Weight​
H: Holes​
HW: Holes Weight​
B: Bumpiness​
BW: Bumpiness Weight

Weight Optimization

1. One-digit brute-force search​

• Searched between 0 and 9 for each weight​
• 10 × 10 × 10 × 10 = 10000 iterations​
• Played 1000 games​
• The total duration is 11 hours ​
• The best score per game is 1513.333​
• Best parameters are 9 | 9 | 9 | 4​

2. Two-digit brute-force search​

• 90 | 90 | 90 | 40 gives same results as 9 | 9 | 9 | 4​
• Searched between -5 and +5 for each weight​
• 10 × 10 × 10 × 10 = 10000 iterations​
• Played 1000 games​
• The total duration was 22 hours ​
• The best score per game is 1589.222​
• Best parameters are 89 | 92 | 89 | 40

3. Three-digit brute-force search

• 890 | 920 | 890 | 400 gives same results as 89 | 92 | 89 | 40​
• Searched between -5 and +5 for each weight​
• 10 × 10 × 10 × 10 = 10000 iterations​
• Played 1000 games​
• The total duration was 26 hours ​
• The best score per game is 1589.222​
• There is no improvement in the best score

Results & Conclusion

• The current average score per game is 1589 for 1,000 games​
• This is a simple solution​
• It can be further optimized using more parameters and different optimization algorithms