griddle makes use of the Simulated Annealing algorithm to optimize and arange grid connections from a state of complete chaos to an optimal state.
griddle is written in Python, tested against 2.7 and uses the following modules:
This is an iterative algorithm, and each cycle will transition the grid to a new state which can be viewed using the pygame interface.
The verbosity can be specified and has the following levels:
- 0 will only print lines on
stdoutfor each iteration. - 1 will display the state of the grid at the beginning and end of the algorithm.
- 2 will display the state of the grid for every iteration - not recommended.
There are 2 important parameters that can be specified:
- Tau is the initial perturbation coefficient - representing in the analogy to simulated annealing how high the temperature us. It's a probability so has to be between 0 and 1, and the bigger the value, the hotter the temperature.
- Beta is the cooling factor, meaning after every iteration this will determine how much we decrease the temperature. A higher number will decrease the temperature more slowly.
For a complete list of all parameters available, you can use the help menu:
$ ./simulated_annealing -h
Usage: simulated_annealing [options]
Options:
-h, --help show this help message and exit
-s SIZE, --size=SIZE size of the grid in format WxH
-t TAU, --tau=TAU initial perturbation coefficient
-b BETA, --beta=BETA cooling factor
-v VERBOSE, --verbose=VERBOSE
verbose level
-n, --nograph
With a verbose level of 1 or more, the initial state will be represented with pygame.
This shows the grid as a set of tiles, and various connections between each of the tiles created at random.
The goal is to minimize the total distance between each tile using the simulated annealing algorithm.
Once the algorithm has converged, the grid will be in a state where the entropy in the system has been minimized with successive temperature decreases. Here is an example of what a grid can look like after the algorithm is done:
Note that it is not guaranteed that the simulated annealing algorithm will find the global optima, as it could also be stuck in a local optima depending on the values of tau and beta chosen - more on that in the analysis section.
You can also specify grids larger than 5x5 with the command line, and the algorithm should work the same way. It is worth noting that as the grid size grows, the tree of probabilities grows as well, and so the algorithm is more likely to get stuck in a local minima, so it is important to decrease the temperature more slowly as you increase the size.
The perturbation coefficient is very important as it controls the accept probability for a new random state. In other terms, it helps avoiding local minima by adding more diversity in the neighboring states, and so should be chosen wisely. In physics term for the analogy to the simulated annealing process, it can be seen as the agitation : if the agitation is too low, we will end up with a local minima, if the agitation is too high, the algorithm will take a very long time to converge.
The following curves show the influence of tau on the energy of the system:
We can see that a higher tau leads to a better exploration of the solution space and is able to converge towards the global minima, while lower values of tau cause it to get trapped in a local minima.
The cooling factor beta is also important because it determines how aggressively the system will cool down after each iteration. It is important not to cool down too quickly otherwise the solution space will not be explored fully, but also cooling down too slowly will be very slow to converge.
