Trying to run multiple simulations of the annealing solver on some data and am struggling to find the best solutions from previous tracks. Are they saved somewhere or do I have to stop the solver right before it moves on to the next one? Thank you!
Hi Sasson,
I used to run Evolutionary Solver(GALAPAGOS & OCTOPUS) and also still in study of that field.
Here is a Brief about Annealing solutions from David Rutten:
you might have seen it.
anyway, just want to chip in this subject.
2 Likes
Thanks @Lei_Yang1. I found the paper while looking for answers to this question. It says the answer from previous runs is cached, but it doesn’t specify where or how to reinstate them. Any ideas on that front?
He @sasson.rafailov did you find the solution to your question? I am having the same issue and can’t get mz head around that problem! It seems so obvious, but I am just to silly, to see the answer.
Best,
Ole
Unfortunately not! I think I just stuck to my earlier plan of stopping the solver as it approached its limit.
Wow! But that can’t be the only solution. @DavidRutten do you have any other idea how to get the results of previous runs when using the simulated annealing solver? Otherwise it could just stop after the one go, right?
Thanks @sasson.rafailov for your quick answer!
Best, Ole
Hi all,
Just bumping the topic to see if anyone found a solution to this? As Ole says, it would seem weird for there to be no means of reinstating solutions from previous annealing tracks - it would surely defeat the point of having multiple tracks if you can’t use them.
I’m hoping there’s something obvious that I’m missing here!
Andy
The point of having multiple tracks is that after a while it becomes impossible for a single tracking to get out of a local optimum. So the solver starts over all the time hoping to find a better solution. When it’s done, it only gives you the best answer is found so far, across all tracks.
I agree that the ability to browse the solutions would be handy, but it’s just not there at the moment.
Apologies, I’ve just realised my mistake!
When I stopped the solver I didn’t realise that the sliders in my script had returned in the background to the ‘all-time’ best configuration - because the list of results on the bottom right of the Galapagos window was still showing only results from the last track, I thought I was stuck with those.
Thanks for your help,
Andy
@DavidRutten The annealing solver has helped me with finding multiple solutions each annealing track.
Next to browsing all solutions I think it would be incredibly useful to safe each result after the solver hits a certain state.
That way we are able to get hundreds of possible solutions and select the one most suited.
Perhaps something for the future.
Definitely part of the future. Once GH2 becomes reasonably useful as a beta product I’ll start anew with Galapagos.
The solution I use is to save a [filtered] continuous log of the slider values and fitness metric used by the solver. Sometimes I save a simple text file, other times I save the data to a database [ MySQL]. When I need to review the results of the solver the data is then loaded back into the system, replacing the slider feeds. I use a stream filter to toggle between sliders and data feeds while reviewing results.
