Question regarding the coupling between problems and files

[chkoar]

Related: openjournals/joss-reviews#2974

Why the problem classes are coupled with files and not with other data structures?

[geoffreyp]

Combinatorial problems implemented (ρMNK-Landscapes, Knapsack and MUBQP) are using instance files. These files are generated with generators (noted in the documentation) by using specific parameters for each instance. I think it will be complicated to implement these generators in our framework to generate exactly the same instance file with the same seed (to compare the results for the same instance between two algorithms with another framework).

Problem classes are strongly inspired by the example code of the generator's authors (for example: http://svn.code.sf.net/p/mocobench/code/trunk/rmnk/generator/java/)

[chkoar]

Since the implemented combinatorial problems are coupled with specific type of problems it should be instantiated using the specific file type and version.

file_mapping = {
    "rmnk_0_2_100_1_0": ...
}


class Rmnk:
    def __init__(file="rmnk_0_2_100_1_0"):
         if isinstance(file, str):
             ....

That way your avoiding the clutter and give the ability to the user to specify the verion of the problem or to pass the file itself is he/she wants to.

[geoffreyp]

Sorry but I am not sure to understand, you propose to replace the str attribute instance_file (for example "/home/mydir/instances/rmnk_0_2_100_1_0.txt") by the attribute file which can be a str (as before with instance_file) or directly a file file=open(instance_file, 'r')?

I don't understand the dict file_mapping but for example for the Rmnk problem, instance files can be named "instance1.txt" or "rmnk1.txt" because parameters are read in the content of the file and not in the file name, so we can't check the filename.

[sjvrijn]

Maybe something like Pooch can be of help? Although that's more aimed at larger files.

[chkoar]

My point is that requiring external files (even for known problems) makes the things difficult. Take a look here. My proposal is to use instance files directly as a last resort. 1) provide the data directly (I know cumbersome), 2) Using a factory and 3)instance files.

[chkoar]

makes the things difficult.

Even for the example in the readme a user is instructed to go somewhere and download a file.

[geoffreyp]

I added all the parameters in problems to allow the user to not use the instance file :
#74

[chkoar]

Let's keep this issue out of the review