I directly ported the TetRandom class to Python, I won't even pretend to understand exactly what the code is doing, but it seems to be a strange variation of a Mersenne Twister. Once Tetris Friends provide a custom seed they can use predictable stream of Tetrominoes in debugging, testing and in multiplayer games.

However, a custom random number generator enables use of their own seed, a facility not provided by the Adobe Math class. The Tetris Friends random number generator was helpfully named 'TetRandom', at first it seemed interesting that Tetris Friends had chosen to implement their own random number generator instead of using the built-in 'Math.Random()'. Tetris Friends games are all Flash based, and as such were easy to decompile using ShowM圜ode.com, some local variable names were lost but the meanings were easy enough to deduce. Tetris Friendsįiguring that Tetris Friends used some form of Psuedorandomness under the hood to refill the Tetromino bags, if I could reimplement the random number generator used in Tetris Friends It would be easy to predict the incoming stream of Tetrominoes given its initial starting conditions. The order of the pieces in a newly generated Tetromino bag should be generated randomly but true randomness is hard to achieve, especially with computers. So, from a given 'Tetromino bag' there are only 7! (5040) different ways in which pieces can be removed. After one of each tetromino has been removed from the bag it is refilled again. It's simple enough to understand, every seven pieces generated during a game are chosen as if they are picked from a bag containing all of the possible pieces. When playing a game of Tetris, the stream of tetrominoes generated by the game typically follows the Random Generator algorithm described in the Tetris guideline. Here in part one of this series I'll describe the process of predicting all the future pieces in a Tetris Friends game and then go into further developments and how to use our new stream of tetrominoes to make gameplay decisions in my next blog post on this topic.

I was curious to see what could be done with a bot if you could predict and generate an infinite stream of the future tetrominoes in a game of Tetris. However when we play Tetris, the game itself imposes a hard limit on how far we are able see into the future, the preview provided is typically only in the range of 1 to 5 pieces. When developing bots for classical games such as Chess and Checkers the depth to which the engine evaluates the game tree is limited by the amount of time and by computing power.

Tetris friends alternative