Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    Luca Cappelletti
    @LucaCappelletti94
    @fserra Well I've implemented a pricer model, what do you mean by "more scip"?
    @mattmilten In my script? Which ones, I'd like to fix them :)
    fserra
    @fserra

    @LucaCappelletti94 you can have a look at https://github.com/SCIP-Interfaces/PySCIPOpt/blob/master/tests/test_pricer.py
    you can implement your pricer within SCIP, so you don't need to have a new model.
    If you are interested, you can learn more about pricers in http://scip.zib.de/doc-5.0.1/html/PRICER.php

    Matthias is referring to the examples that we have in PySCIPOpt. We've been lazy because we do not really have example written the scip-way. However, the tests, they are the scip-way

    Luca Cappelletti
    @LucaCappelletti94
    Ok, thanks! I'll take a look!
    May I ask a couple questions about the script? I studied Operations research in Italian so I have some difficulties with initials and specific terms.
    Matthias Miltenberger
    @mattmilten
    Please excuse my late reply. Of course you can ask us about the code. Go ahead!
    MG
    @Cerebrock
    hi
    i have like 1.400.000 variables for Constr opt... i'm screwed right?
    fserra
    @fserra
    maybe.
    did you try solving? what happened?
    MG
    @Cerebrock
    i'm using cvxpy, haven't tried scip yet but it's my hope. It would be very nice to have an interface
    Matthias Miltenberger
    @mattmilten
    Just as @fserra said: go ahead and give it a try! And tell us about your experience :-)
    MG
    @Cerebrock
    I was hoping someone here had a nice lil utility function lol 😋
    Matthias Miltenberger
    @mattmilten
    Ah, now I understood. I haven't heard of such a converter but I guess it shouldn't be too hard to build up a model in PySCIPOpt based off of you cvxpy code. If you only want to see how SCIP can solve the model you might want to write out the cvxpy model as LP or MPS file and feed it into SCIP.
    Luca Cappelletti
    @LucaCappelletti94
    @mattmilten sorry for disappearing for so long, I was dead in the water with my thesis. I wanted to ask you if it could be possible to setup a hangout call to update the column generation example together so that it is fully in a pysciopt way. I suppose it would take at most one hour.
    Matthias Miltenberger
    @mattmilten
    Sounds good! I am quite busy at the moment, but maybe @fserra or @stephenjmaher could help out?
    Luca Cappelletti
    @LucaCappelletti94
    Do let me know!
    stephenjmaher
    @stephenjmaher
    @LucaCappelletti94 I am happy to talk to you about a column generation example. Can we set up a call Tuesday next week?
    Luca Cappelletti
    @LucaCappelletti94
    Tuesday the 22th?
    stephenjmaher
    @stephenjmaher
    Yes, the 22nd.
    Luca Cappelletti
    @LucaCappelletti94
    @stephenjmaher Would it be okay at 9GMT?
    Viech
    @Viech
    Hello, does anyone know whether the barrier convergence tolerance option bounds an absolute or a relative deviation from the optimum value?
    @Cerebrock PICOS is a high level interface similar to CVXPY that supports PySCIPOpt, if I may drop the advertisement. :)
    Chitrang Srivastava
    @c0D3M
    Hi, I am newbee and using pyscipopt, so far works good, but how can I enable LNS (Large Neighborhood search), becuase for larger instances it took forever
    Matthias Miltenberger
    @mattmilten
    Do you mean "disable"? In any case, you can play around with the frequency of heuristics if they are taking too much time. -1 for disabling a heuristics and 1 for enabling it at every node. You can change settings using setIntParam("param name", value).
    see here for a list of all available parameters: https://scip.zib.de/doc/html/PARAMETERS.php
    Chitrang Srivastava
    @c0D3M
    Thanks, let me try
    CGudapati
    @CGudapati
    I am not sure if this room is still active as the last message is in March! But I will try my luck either way. Are there any examples of branch and price in pySCIPOpt? I looked at the tutorials and the most promising was the bin packing but the colgen is still a heuristic.
    stephenjmaher
    @stephenjmaher
    Unfortunately, there are no examples on exact branch-and-price. I am happy to talk you through the required implementation steps if you would like.
    CGudapati
    @CGudapati
    Thanks for your help. My problem has two variables z and x. I only need to branch on z variables. What I had in mind was to perform diving and get a good bound and then I just need SCIP to branch on z variables (setting them to 0/1) and branch.
    stephenjmaher
    @stephenjmaher
    Your question seems to be more in general about how to use SCIP. Do you have any specific branch-and-price questions? If you only want to branch on z, then you just need to set x to continuous and z to binary.
    CGudapati
    @CGudapati
    I might not have been very clear. I mean it would’ve great to see an example where I can branch and then add more columns.
    For instance I followed this guide for column generation https://scipbook.readthedocs.io/en/latest/bpp.html
    It ends with just column generation and there is no branch and price.
    It just solves the model as am IP in the end.
    CGudapati
    @CGudapati
    for my problem, after adding all the columns, If I solve it as a simple IP, I won’t be getting the optimization solution. But by branching and further adding columns, I might get optimal solution. I want to know if there are any notes or examples which shows branching and adding more columns using an external pricer
    This is only code I could find
    But
    It
    seems a bit dated and still experimental.
    stephenjmaher
    @stephenjmaher
    If you want to perform branching throughout the tree you need to add a Pricer object. A Pricer object include the member function pricerredcost that is called when pricing is to be performed when solving the LP of a node. An example of how to implement a pricer object is in tests/test_pricer.py in the PySCIPOpt distribution.
    CGudapati
    @CGudapati
    I apologize for not being very clear. I have already implemented my Column Generation Code. Hopefully I am being more descriptive now.
    1) I have created my LP master problem.
    2) I solve it, get dual prices, then use these dual prices in my pricing problem (it's a variation of shortest path problem) to add a new column.
    3) I do it till I can not add any more columns.
    4) Now, I want to branch on some variables. When I set a variable to 1, I can "solve" that branch and repeat the steps 2-3 to get new column. I am failing at STEP 4). i.e how to declare branching rules and add add more columns after setting a variable to 1 (I do not care about setting to 0)
    I most likely will move to C++ as there are more examples. Thanks for your help :)
    Sergey Bartunov
    @sbos_twitter
    Hi everyone! Is there a simple way of getting an LP-relaxed solution of an integer program without manually changing type of each variable to continuous and re-solving the problem?
    Matthias Miltenberger
    @mattmilten
    No, I suppose that is the most simple way to do it.
    Nitish Joshi
    @nitishjoshi25
    Hello, Im trying to use SCIP with CVXPY. CVXPY's installation page asks to install PySCIPOpt for this..
    however CVXPY is still not able to 'find' the SCIP solver
    has anyone faced a similar problem ?
    Nitish Joshi
    @nitishjoshi25
    installed the newest version and the problem was solved