getCallbackSolution failing and yielding error 422 inside a chgbranchobject callback?

According to the documentation getCallbackSolution() from within a chgbranchobject callback should indeed return the current relaxation.

I cannot reproduce your problem, though. I tried several problems with Xpress 45.1.2 and called getCallbackSolution() for each and every invocation of the callback. I did not observe any problem.

What version of Xpress are you using? What object are you calling getCallbackSolution() on? You must call it on the problem object that is passed as first argument into the callback, not on the problem object on which you called optimize(). Can you share your code or provide a short example that reproduces the problem?

Can you please also show the rest of your code (how does your callback look like, how do you register the callback, how do call optimize())?

What version of Xpress do you use?

In order to reproduce the problem here, we would also need the CSV file with your data.

Thank you Daniel for your reply and sorry for my laconic one,unfortunately I got my account blocked for whatever reason as I tried to attach multiple screenshots.This time I'll paste the code as plain text,for good measure.After the model definition we have:

class controldata:

 def __init__(self,inputdata,outputdata,bestbound,nodenumber,brancher,threshold,threshindex,Mleft,Mright):

   self.bestbound=bestbound

   self.inputdata=inputdata

   self.outputdata=outputdata

   self.nodenumber=nodenumber

   self.brancher=brancher

   self.threshold=threshold

   self.threshindex=threshindex

   self.Mleft=Mleft

   self.Mright=Mright

controlmatrix=controldata(X,y,10000,[],[],[],[],[],[])

(An object to be passed to the callback,useful only for providing the input and output samples X and y,most of the attribute lists were meant for an early implementation and will stay empty)

def spatialbranch(prob,obj,obranch):

  N=np.shape(obj.inputdata)[0]

  X=(obj.inputdata)

  y=(obj.outputdata)

  h=math.floor(0.8*N)

  sol=prob.getCallbackSolution(z)

(The actual callback,up to the point where the issue is encountered.We might argue that z is not defined inside the function,but only as an xpress object of the prob we'll pass as input argument;still,even using getCallbackSolution() with no argument inside the brackets to get the full solution will result in the execution stopping with error 422).

I add the callback to the problem with p.addcbchgbranchobject(spatialbranch,controlmatrix,5),and I then optimize with p.mipoptimize().

As for the CSV file,it's one of several regression datasets available on Kaggle as part of the Starter: R robustbase Datasets 6f223ec8-4

package;I doubt the issue depends on how the CSV's are parsed,since optimization of the problem on the same dataset without custom branching will work fine.

Also,my Xpress version is 45.01.02,as per the output of the xpress.getversion() method.

Thank you for your code. We have reproduced the problem here and are investigating. We will come back once we found the problem.

For the next time: it is best to share code either as attachment or as text. That makes it easier for us to copy/paste and run the code here. Not sure why you got blocked for uploading some screenshots, though. Sorry about that :-(

Thank you,what has been bugging me in the meantime is,do you happen to encounter the same issue,feeding that specific model to the callback?(Notwithstanding what else would be inside the callback,even using a dummy function returning obranch after the sol=getCallbackSolution() )

Yes, we used your exact problem.

As far we can tell at the moment, this is a problem with the strong branch initialization for problems with quadratic objective.

One way to avoid this to disable strong branching and skip the callback for the root node. In order to that, do

p.controls.sbbest=0

before calling mipoptimize() and in the callback do

  if prob.attributes.nodes < 1:
    return

That fixed the problem for me (I did not investigate the performance impact of disabling strong branching on the first node)