gurobi solution modeling #1374
Replies: 23 comments 21 replies
-
|
So how do you know the solution isn't optimal? And what do you mean with gurobi+. If you selected 'gurobi' as solver, and yalmip told you it cannot use 'gurobi', and you then selected 'gurobi+', it just means you asked yalmip to create a nonsense model for gurobi, and thus a nonsense solution will be returned |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
gurobi cannot deal with that, it is not SOCP-representble (unless x or y is integer in which case it can be massaged into something MILP-representable) |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
If x or y is integer it can be rewritten using MILP. If not, it cannot be solved using Gurobi unless approximated NEVER use the + operator to select solver. It means "Hey YALMIP, you are incorrectly claiming that gurobi is not capable of solving this problem. I know better than you so I force you to setup a model which you claim gurobi cannot handle, and send that model to gurobi, and then give me back what ever it returns, despite it telling me that it cannot solve the problem" |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
MILP is mixed-integer linear programming. If x or y are binary or integer, you simply model it using implications https://yalmip.github.io/tutorial/logicprogramming#:~:text=y%20%3D%20f(x)%2C%20x%20scalar%20integer Basically "if x = 0 then obj = y, if x=1 then obj = 2*y" etc |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
Integers are just more implications. 2^xy is simply "if x = 0 then obj = y, if x = 1, then obj = 2y, if x = 3 then obj = 8*y" etc |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
Yes, using implies and possibly new variables to massage things to something MILP (or MIQP)-representable |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
I have no idea what you are talking about |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
You will have to come up with a way to model your situation without using variable whose structure depends on another variable. It is simply not possible to model it that way. Typically, you will have to use a variable whose length is the longest possible, and then model using that, and impose stuff to represent the stuff that you really want to happen when the length is only k (forcing remaining variables to be zero etc) |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
Unclear what you mean, but maybe this https://yalmip.github.io/tutorial/logicprogramming#:~:text=much%20as%20possible-,Counters%20and%20loops,-A%20very%20common |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
I have no idea what you are asking about |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
you're not showing the error |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
exponentials have to be approximated https://yalmip.github.io/command/interp1/#:~:text=Example%20(Linear%20optimization) |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
You cannot use decision variables in programming constructs (for-loops etc). Everything has to be written using linear operations or high-level logic conditions implemented using implies (i.e. mixed-integer linear operations) |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
what would break even mean without there being a for (or while) loop somewhere. break what? |
Beta Was this translation helpful? Give feedback.
-
|
When calling gurobi in yalmip for solution, the objective function contains (2^x)*y, and x and y here are intvar sdpvar variables respectively. If you do not use implication, how to deal with approximation, is there any example
|
Beta Was this translation helpful? Give feedback.
-
|
since you're already in the integer programming world, I don't see why you would want to deal with this using some crappy approximation, as it is trivial to represent exactly |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for your reply, because the data of integer x at this time is a little huge, if you use implication to find out the minimum need to simulate several times, so I want to approximate an optimal solution
Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
no idea what you mean with simulate several times. a simple binary model would be the easiest. if you maximize (2^x)y it can be represented with a complicated geomean model. I don't see an approximation being performant |
Beta Was this translation helpful? Give feedback.
-
|
implies is used in the model, and Gurobi is chosen to solve it. From the output lp file, implies that the default M value is 10000, is there a way to correct this value
Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
If YALMIP has resorted to the bound 10^4 it means you are not bounding variables sufficiently well or you have bad scaling in your model. If you against my advice would want to force a larger M and thus create even more numerically unstable and weak relaxations you will have to edit derivebounds.m |
Beta Was this translation helpful? Give feedback.
-
|
Thank you, Sir. I opened the derivebounds.m file. Do I change the 1e4 Settings here.----------
M(isnan(M)) = inf;
M(isinf(M)) = 1e4; M = M+b;%+0.01;
m(isnan(m)) = -inf;
m(isinf(m)) = -1e4;m = m+b;%-0.01;
|
Beta Was this translation helpful? Give feedback.
-
|
yes But as I said, you should improve your model so sensible bounds can be derived and not force a specific M |
Beta Was this translation helpful? Give feedback.
-
|
Hello teacher, I have two models, one is a general model, whose optimization variables are represented by intvar, for verification; the other is a deterministic model, whose corresponding variables are input by the optimization result value of the first general model intvar variable, all of which call gurobi, and the solver parameters and other information of both are consistent. The difference is only that the value of a variable is intvar, and the value of a variable is input with the value of the optimization. The optimal value of the output of the two variables is all the same, but there are differences in some decision variables, which will not affect the target result. Can we consider these two models consistent, or what causes them
|
Beta Was this translation helpful? Give feedback.
-
|
aren't you simply saying you have a problem with a non-unique optimal solution |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for your reply, do you mean the case of multiple solutions? It is indeed a bit like the case of multiple solutions. The difference between the two models is only that, in order to verify the general model, the output result of the general model optimization variable intvar is directly input to the deterministic model as the value double. The rest is the same, the objective function of both is the same. There are small differences in the remaining decision variables that run but have no effect on the objective function, or how to determine that this difference is due to multiple solutions rather than differences in optimization using intvar and double modeling respectively.
|
Beta Was this translation helpful? Give feedback.
-
|
no one will know the answer to this better than you. you use two different models, and get the same solution on the important part but different on other. if you think that is fine, well that is up for you to decide |
Beta Was this translation helpful? Give feedback.
-
|
Hello, the default implies in yalmip that the large M is 10000, and what about the variable, is there a limit on optimizing the variable
…------------------ 原始邮件 ------------------
发件人: "wnn2201" ***@***.***>;
发送时间: 2024年4月13日(星期六) 下午3:44
***@***.***>;
主题: 回复： [yalmip/YALMIP] gurobi solution modeling (Discussion #1374)
Thank you, Sir. I opened the derivebounds.m file. Do I change the 1e4 Settings here.----------
M(isnan(M)) = inf;
M(isinf(M)) = 1e4; M = M+b;%+0.01;
m(isnan(m)) = -inf;
m(isinf(m)) = -1e4;m = m+b;%-0.01;
|
Beta Was this translation helpful? Give feedback.
-
|
? |
Beta Was this translation helpful? Give feedback.
-
|
Teacher if the variables don't currently have upper and lower bounds set on them, then yalmip's default bounds on variables are 1e4 in derivebounds, which means the maximum value of the variable can only be taken to 1e4 here
wn
***@***.***
…------------------ 原始邮件 ------------------
发件人: "yalmip/YALMIP" ***@***.***>;
发送时间: 2024年4月13日(星期六) 下午3:35
***@***.***>;
***@***.******@***.***>;
主题: Re: [yalmip/YALMIP] gurobi solution modeling (Discussion #1374)
If YALMIP has resorted to the bound 10^4 it means you are not bounding variables sufficiently well or you have bad scaling in your model. If you against my advice would want to force a larger M and thus create even more numerically unstable and weak relaxations you will have to edit derivebounds.m
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
Teacher if the variables don't currently have upper and lower bounds set on them, then yalmip's default bounds on variables are 1e4 in derivebounds, which means the maximum value of the variable can only be taken to 1e4 here，right?
wn
***@***.***
…------------------ 原始邮件 ------------------
发件人: "yalmip/YALMIP" ***@***.***>;
发送时间: 2024年4月13日(星期六) 下午3:35
***@***.***>;
***@***.******@***.***>;
主题: Re: [yalmip/YALMIP] gurobi solution modeling (Discussion #1374)
If YALMIP has resorted to the bound 10^4 it means you are not bounding variables sufficiently well or you have bad scaling in your model. If you against my advice would want to force a larger M and thus create even more numerically unstable and weak relaxations you will have to edit derivebounds.m
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
Yes, and if you cannot improve your model you will have to hard-code a change in derivebounds.m |
Beta Was this translation helpful? Give feedback.
-
The solution found by using gurobi+ is not optimal, the output of decision variables is all upper limit, the yalmip version is 2021.03.31, and the gurobi version is 10.0.3. My problem is the same as the problem discussed before when the time is 2023,04,21. After turning off presolve, the result remains unchanged. How to solve this problem, and the installation of yalmip version and gurobi version has anything to do with it?
Beta Was this translation helpful? Give feedback.
All reactions