StackOverflow2013

Note that there are some explanatory texts on larger screens.

plurals

PO
text
Body
copied!<p>Here is <a href="http://www.lalena.com/AI/Tsp/" rel="nofollow noreferrer"><strong>a C# program</strong></a> approach for what you are looking for.</p> <p>With regards to the interest (or lack thereof) of implementing <strong>cross-over</strong>, it all hinges on the particular selection logic your implementation will use (and/or the evaluation function itself, if for example it includes an evaluation of the speed of improvement). In many cases, <strong>cross-over operations will "rescue from the chopping block" some solutions that are effective/optimal in an area of the graph but somehow "stuck" in others</strong>. This is not to say that if the overall algorithm is slow-enough and covers a good percentage of the solution space the same solutions may not have been discovered anew, but cross-over may also increase these discoveries (and/or letting you stuck in another local minima ;-) )</p> <p>Not directly related but of noteworthy interest to whomever looks into GA, is the <a href="http://www.usc.edu/dept/molecular-science/papers/fp-sci94.pdf" rel="nofollow noreferrer"><strong>original "ultimate" experiment in GA</strong></a> original "ultimate" experiment in GA by Prof. Alderman (of RSA fame), who used actual DNA molecules [into a C program - just kidding] to solve a related graph problem, that of hamiltonian graphs.</p> <p><strong>Edit</strong>: In re-reading the question I understand why you asked it or more precisely <strong>why you'd like a "No you don't want cross-over" reply</strong> ;-)<br> Your <strong>genonme is directly tied to the graph</strong> itself (nothing wrong with that, <em>a priori</em>), but this brings the impediment that most cross-over offsrpings will not be viable, since they may may have duplicate nodes (visit same city twice or more) and be missing nodes (failing to visit some cities)... Furthermore, viable cross-overs will affect similar graphs, and hence maybe merely incrementally expend the search, compared with what mutations would have discovered...<br> Hum... Then maybe <strong>cross-over, <em>in this particular implementation</em> won't help the algorithm very much</strong> (and indeed take much of its CPU to create, test and often discard cross-over offsprings, CPU which would be better used by affording <strong>more iterations, and a slower cooling rate</strong>...). Unless! You find a clever way of performing cross-over operatitions ;-)</p>

Querying!

Guidance

An individual column

Larger individual text columns get their own page to allow for proper reading.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload