#Bin-Packing CSCI-338 Bin packing assignment
This problem is much more closely related to the strip packing problem. If one were to, say, want to google keywords for ideas, that might return better results.
##Goal
Place a list of rectangles into the smallest bounding box (measured by perimeter, not area) you can manage, as fast as possible, without overlapping. TLDR at the bottom.
##Why This problem, and its variants are probably NP-hard. The NP-Complete variant of this problem can be stated as:
"Given a set of rectangles, can they be fit into an area less than or equal to some area x?".
Because the problem we are providing has a lot more structure than some of the other NP-hard problems, it is actually possible that you could exploit the structure to solve this quickly, but I somewhat doubt it.
The purpose of this assignment is to display the practical concerns of solving difficult and intractable problems.
This problem is legit an actual real world problem.
##How Invent some strategy to hopefully do better than some awful solution.
The simplest solution would be placing each rectangle in a single, long, row. This would be faster than any other solution, but would provide a reasonably awful solution most times.
Most of us could easily find solutions that are somewhat okay. Try to figure out how you are capable of doing this; and make a program to do so.
##You need This code! See bottom for explanation of code stuffs.
##You provide You will create a Python module with a function named "find_solution" that accepts a single argument consisting of a list of tuples of type (int, int) representing the (width, height) of n rectangles. Your function will return a list of tuples of type (int, int), specifying the (x,y) positions of the rectangles. See sample_solution.py for an example.
I will run your code with the code provided, you will ensure your program can be successfully called by mine. You can look at the example to see exactly how that works. I'll be using standard data sets for grading.
Your code will return a list of positions, as tuples, that represent the positions of the rectangles passed in. The nth tuple you return is the position of the nth rectangle provided.
Because of this you cannot rotate rectangles.
We will validate your results to ensure that no overlapping occurs, otherwise it is not a valid solution.
You are both encouraged and expected to test your solution with the provided code. ##Grading
We can't grade solely on speed, because that incentivizes terrible solutions.
We can't grade solely on smallest area, because that incentivizes taking till the end of time finding the optimal solution.
Therefore, we must factor both area and time.
- We will run your code on some number of data files.
- Each data file will have a maximum allowed time.
- You pass a data file if the time taken to generate a solution is less than the max time.
- The way we will determine the quality of your solution is by comparing it to the rest of the class's solutions.
- You also fail a set if any of your rectangles are overlapping.
What does this mean?
This means that if you solve every data file with awful solutions, you get a bad score.
If you solve only 1 of them with a fantastic solution, you get a bad score.
We want you to pass a bunch of them with pretty good solutions.
The reason for this is that it is realistic. If you were expected to solve this for a job, they don't want to you to take 100 million years to get a really good solution, nor do they probably want you to give them the worst answer possible right now.
The specifics for grading are on Paxton's site.
##TL;DR
- Give each of n rectangles an intenger position, such that they do no overlap.
- Do it fast.
- Pack it as tightly as possible.
- You want to pass as many data sets as possible, this is the most important thing.
- In the case that someone solves as many as you, whoever has the smaller solutions wins.
- To test a solution, you run driver.py
- It will ask for the name of one module with a solution, and then for another. Then it will ask for the number of datasets to run on.
- driver.py has default values for loading modules and the number of datasets to use.
- Just pressing enter will run these defaults.
- If you feel the need to have different defaults, they are defined at the top of driver.py.
- If you have pyglet installed for python 3, you will be prompted for if you want to display a graphical representation of the two solutions' packings.
- Getting the visualizer working is in the next section.
- I find it very useful for figuring out why solutions are not functioning as intended.
- Currently, the values for rectangle sizes are fixed, I will be allowing you to change them probably, you can edit them in the meantime if you care about that. That is located at the bottom of driver.py
- You don't need this, it's just for fun.
- The only dependency for the visualizer is Pyglet for python 3.
- Pyglet has no dependencies of its own, so a nice "sudo pip3 install pyglet" will work for Mac (and possibly linux users).
- For Windows users 👎, pip is located in the install directory of python. If you can run command prompt in that directory, you too can install pip.