Statistics

Now I have some data to show and I made some cool graphs. First, I would like to explain the meaning of the terms "Round" and "Try" in my statistics. When someone plays Pacman and die 3 times, it is one round, and his/her first try. If he/she keep playing, without closing the game, it is going to be the second try, and in that way go on. Second, I ran the QLearning Pacman for 4600 rounds, the Random Pacman for 2000 rounds, and 4 different humans played two rounds each. It is a pretty small number of human players, but we can have some general view from that anyway. Lets go to the graphs:

So, when we are playing against each other, the most common question is "What is your high-score?" than, I decided to compare it, and the result was pretty impressive. The Qlearning algorithm achieved a score almost two times higher than the humans.

Someone could say: "Okay, but I am sure that humans gonna have a better average!", but the data shows the opposite. Even with a smaller difference, the Qlearning algorithm still showing a better score.

At first, I thought that I could see the evolution of the Pacman through his tries, so I build that graph bellow that shows the average of scores through his tries. Otherwise, the results showed that the learning of the algorithm is so fast, that it is not really visible through tries, because it happen just in the first moment of each experience. In this chart I decided to use a average of all human scores to be seen as a reference of performance.

QLearning Pacman is better than humans?

I wouldn't say that, besides the fact that we have a really small population of human rounds, the way that I designed the game can influence in the final result of that data. I designed the game to increase the difficulty exponentially through the levels intended in make the game quick, interesting and challenger. As we can see in the graph bellow, the humans has a tendency to try go on through levels faster that the QL Pacman. It happens because I gave a high reward to the QL Pacman when he ate ghosts, so he decided to give priority to that instead of only go to the next level. Otherwise, it doesn't mean that the humans use a bad strategy, the key point here is that the level 2 (which is the third) is really hard and no one could get through it.