Invalid variable index #489
Comments
|
I need to reproduce the bug on my computer to fix it. |
|
Hi, Guillaume. You need to run with the Coluna parameters set as above in file run.jl of CapacitatedLotSizing application. To run, you can use: julia src/CapacitatedLotSizing/scripts/runsingleinstance.jl X11119APlease note it will take a few minutes to get to this error message. I am not sure, but I guess this happens when the algorithm reaches some stopping criterium, such as node limit, time limit etc. Also, I get the same error if using the default depth first strategy. |
|
I could reproduce the bug, thank you. |
|
Ok I think I spotted the bug : In node 240 (parent 239), A column Next node explored is 241 (parent 239), The subsolver is synchronized. We note that the variable Fix : do not update cost |
|
@rrsadykov should we flush all buffers at the end of the each node ? |
|
I do not understand what is "subsolver", and also what is the "buffer" and what does "flush all buffers" means |
|
|
Normally, when passing from node to a node, the state of the master columns storage unit should be restored, thus the column in question should be deleted. Is the bug in this part? For me, as the column is not present in the formulation, the storage unit is restored correctly. Unfortunately, I do not control the buffer part, as I did not implement it. I do not understand why the cost of the column is updated if this variable is not present in the formulation. It is not marked as non-active? |
|
Yes it's normal that you don't control the buffer part. The buffer should automatically store changes done to the formulation. In the buffer, variables added to the formulation and changes made on the cost of the variable are stored separately (fields Coluna.jl/src/MathProg/buffer.jl Lines 48 to 59 in 43c44a4 When one deactivates a variable, the variable is removed from the So I think it's time to start unit tests on the buffer. To reproduce this bug :
|
Hi @guimarqu . Did you reached the error using the default pricing solver by MIP in Coluna or the one in Please consider to use the default from Coluna since I spotted a bug in my pricing callback that may lead to the addition of wrong columns. I will create a new merge request with the fixing soon. This is the output using the default pricing MIP solver at the same point you reached the bug: <it= 2> <et=273.81> <mst= 0.05> <sp= 0.09> <cols= 0> <al= 0.00> <DB=28925.9000> <mlp=28925.9000> <PB=Inf>
[ Info: Column generation algorithm has converged.
[ Info: No branching candidates found. No children will be generated.
***************************************************************************************
**** BaB tree node N° 239, parent N° 233, depth 125, 121 open nodes
**** Local DB = 28096.5641, global bounds : [ 26871.7802 , Inf ], time = 273.82 sec.
**** Branching constraint: v<=0.0
***************************************************************************************
<it= 1> <et=274.10> <mst= 0.12> <sp= 0.10> <cols= 5> <al= 0.00> <DB=28096.5641> <mlp=28110.7174> <PB=Inf>
<it= 2> <et=274.27> <mst= 0.04> <sp= 0.12> <cols= 3> <al= 0.00> <DB=28096.5641> <mlp=28110.7174> <PB=Inf>
<it= 3> <et=274.40> <mst= 0.04> <sp= 0.09> <cols= 1> <al= 0.00> <DB=28096.5641> <mlp=28110.7174> <PB=Inf>
<it= 4> <et=274.52> <mst= 0.02> <sp= 0.09> <cols= 0> <al= 0.00> <DB=28110.7174> <mlp=28110.7174> <PB=Inf>
[ Info: Column generation algorithm has converged.
***************************************************************************************
**** BaB tree node N° 240, parent N° 239, depth 126, 122 open nodes
**** Local DB = 28110.7174, global bounds : [ 26871.7802 , Inf ], time = 274.54 sec.
**** Branching constraint: v>=1.0
***************************************************************************************
<it= 1> <et=274.68> <mst= 0.03> <sp= 0.09> <cols= 0> <al= 0.00> <DB=28111.3429> <mlp=28111.3429> <PB=Inf>
[ Info: Column generation algorithm has converged.
***************************************************************************************
**** BaB tree node N° 241, parent N° 240, depth 127, 123 open nodes
**** Local DB = 28111.3429, global bounds : [ 26871.7802 , Inf ], time = 274.71 sec.
**** Branching constraint: v>=1.0
***************************************************************************************
<it= 1> <et=274.87> <mst= 0.05> <sp= 0.09> <cols= 0> <al= 0.00> <DB=28117.9000> <mlp=28117.9000> <PB=Inf>
[ Info: Column generation algorithm has converged.
[ Info: No branching candidates found. No children will be generated. |
Ok, then one just needs to remove it from |
Yes and test because we might have other hidden bugs. |
|
@hhkramer The bug is independant from the pricing solver so the fix will work for both cases. |
|
OK, thanks @guimarqu ! Just as an update, after fixing the bug in my pricing callback I could not reach the error again. It is running for more than an hour and 1500+ nodes were solved. |
Describe the bug
When using Coluna to solve my application I get the following error after some nodes of the branch-and-price tree:
ERROR: LoadError: The index MathOptInterface.VariableIndex(-1) is invalid. Note that an index becomes invalid after it has been deleted.I am using the default
TreeSearchAlgorithmparams except that Primal Heuristics are disabled.To Reproduce
Coluna parameters are set as follows:
Expected behavior
Environment:
The text was updated successfully, but these errors were encountered: