The swarmbot competes with the bots on eOthello. These Graphs show the percentage of wins against an opponent.
Monte Carlo Agent
Ant Colony
Firefly
Cuckoo Bird
These metrics were collected by timing how long it takes each swarm to perform 3000 dives in parallel. The Monte Carlo graph is the only one where you will see performance on one agent. The Ant Colony, Firefly, and Cuckoo Bird swarms are reliant on more than one agent, so there is no measurement for a single agent. One thing to note is that the Ant Colony graph is measured in hundredths of a second compared to the rest of the swarms performing at tenths of a second. Even two Ant Colony agents can outperform eight Monte Carlo agents.
Left click or, if on a phone, tap on a square to go down the tree. Right click or long press on a square to go back up the tree. The number before the dash is not significant to the gametree. The google charts API just needs a unique id to build the tree properly. After the dash represents the move being explored. These graphs show the behavior of how the different swarms explore a gametree. The squares that are green show a node that points towards more winning strategies. Red nodes point towards less winning strategies. If a graph starts with one node, that is a game where the agent plays second, so the first part of the tree was chosen by the opponenent.
Note: These tree maps only go about ten levels down the tree. With a state space of 1028 the full gametrees are very large and trying to put four of them on a website would take forever to load up. Ten levels down gives between 4000-7000 nodes to look at and gives a good representation of how the agents explore the gametree.
You can see in the Monte Carlo tree map that the colors are faded and get more defined as you travel down the tree. This shows a more random dispersement when searching the tree. Whether There was a win or a loss doesn't matter to the agent until the simulations are over and it is time to choose a move.
In the Ant Colony graph, we can see a targeted approch to searching the tree. The pheromone trails that the ants follow allow the swarm to favor the winning paths. Since there is still a probabilty of veering off the pheromone trail, the agents can still explore around the pheromone trail to find more wins.
The Firefly graph shows a similar start to the Monte Carlo graph where the colors are faded showing a random dispersement. This is expected since the Firefly and Cuckoo Bird swarms start out in the same manner as the Monte Carlo agents. As the genetic algorithms start to mutate and favor agents that find more winning strategies, The colors get more defined at a quicker pace compared to the Monte Carlo.
The Cuckoo Bird graph shows a similar exploration approch to the Firefly. Colors get more defined and targeted as the Genetic algorithms start favoring winning strategies.