Speed up the solver() about riskfolio-lib

[dcajasn]

Hi @lemon234071,

It depends on the solver that you choose. To reduce computation time is better to use a commercial solver like GUROBI, CEPLEX or MOSEK. All of them offer academic licenses, but personally I prefer MOSEK because its academic license is less restrictive and also supports all kind of cones. On the other hand, for linear models GUROBI is a little faster in academic version but has more restrictions in the number of variables and constraints.

Best,
Dany

[lemon234071]

Hi @lemon234071,

It depends on the solver that you choose. To reduce computation time is better to use a commercial solver like GUROBI, CEPLEX or MOSEK. All of them offer academic licenses, but personally I prefer MOSEK because its academic license is less restrictive and also supports all kind of cones. On the other hand, for linear models GUROBI is a little faster in academic version but has more restrictions in the number of variables and constraints.

Best, Dany

I found that the rp.Portfolio() has the default self.solvers = ["ECOS", "SCS", "OSQP", "CVXOPT"].
So, all i need to do is change it like self.solvers = ["MOSEK"] ? Will there any hidden danger？

[dcajasn]

I use these solvers by default because are open source and are installed by default with CVXPY.
Yes, that's the idea but first you must install MOSEK and get an academic license in its website. If you see Entropic Value at Risk tutorial, I use MOSEK to reduce computation time.

[lemon234071]

Thank you first, i just tried MOSEK with just pip install MOSEK . It faster but it return the None solution while other solvers return a weights....

[dcajasn]

It's because you need a valid license. In riskfolio-lib install instructions are the links to get an academic license.

[lemon234071]

Thank you very much i ll try it.

[lemon234071]

It works thank you very much, btw, did you try the Cvxpylayers on GPU? Would it be faster???

[dcajasn]

No, I didn't try cvxlayers. I use a Macbook air, so I don't have a GPU to try cvxlayers, but based on the information I read on papers, blogs and forums; commercial solvers like GUROBI, CPLEX, XPRESS and MOSEK are the best.

[lemon234071]

Thank you very much！

[lemon234071]

Hello there, me again! Thank you for your time first!

I still have some problems with the speed:

1, With constraints "upperlng“ and “nea”， my problem should be solved with numerical methods.
Do the different objective functions heavily affect the speed?
(i mean I saw that optimization of sharpe is realized by minimizing the risk with constraints. I mean do the different forms of the function with constraints will be solved by different methods detected by mosek?).
2, Do the parameters of Mosek affect the speed and the performance? (i mean in self.sol_params)

My sincerest thanks！
Best wishes!

Some info with cvxpy

[dcajasn]

1. Yes
2. Yes