Getting started¶

This framework can be used both as a command line tool and as a python module. When using as a command line tool use the -h flag to see all available options:

fastlane -h

The framework currently supports implementation of the clingo, clingo-dl and clingcon solvers, and the approaches of LNS, ALNS, LNPS and ALNPS, selectable via different options and presets.

For finer control one can implement their own solver and/or strategy or modify existing ones through inheritance.

Any solver or other component should be implemented according to the interfaces described here and has to implement all abstract methods. More details to the concrete implementation can be found in the corresponding section.

When using the framework as a python module, the LNS object has to be initialized with the encoding files. A dictionary of parameters can be passed during initialization of the LNS object, which intern is used to overwrite attributes of the LNSOptions object. Alternatively, already initialized LNSOptions objects can also be provided.

The 'preset' parameter can be used to set multiple parameters at once according to predefined presets. Manually set parameters will always override preset values. And preset values will overwrite default configuration values. An example of the LNS initialization can be seen below or in ./examples/demo.py. For a more detailed and step by step introduction to the framework, check out the guide.

from fastlane.lns import LNS
from fastlane.lns_options import LNS

options = LNSOptions(seed=42)
lns = LNS(
    ["./examples/golf.lp"],         # ASP encoding
    {                               # set additional parameters
        "time_limit": 60,           # overall time limit in seconds
    }
    options,                        # LNSOptions object containing defaults
)

The same search can be performed through the command line as follows:

fastlane --seed=42 --time-limit=60 ./examples/golf.lp 

Encodings¶

The encodings should contain some kind of optimization statement or soft constraint. The lns framework will work with the solution cost derived by the solver.

When using --destruction=declarative, additional helper atoms must be specified. These include:

_project_op(ID,S).       % ID: identifier, S: signature 
_project(ID,A).          % ID: identifier, A: affected atom
_destroy_op(ID,V)        % ID: identifier, V: value i.e. percent/number/auto
_destroy(ID,A,S)         % ID: identifier, A: affected atom, S: selected term for destruction
_prioritize_op(ID,V,M)   % ID: identifier, V: value, M: modifier
_prioritize(ID,A).       % ID: identifier, A: affected atom
_config(ID,PID,DID,PIID) % ID: identifier, PID: project id, DID: destroy id, PIID: prioritize id
_strategy(ID,CID)        % ID: identifier, CID: config id

Benchmark-Tool¶

The benchmark-tool section provides some guidelines and examples on how to use this framework with the potassco-benchmark-tool.