Monte carlo simulation optimisation framework C#
- Thread starter John TRD
- Start date
Daniel Duffy
C++ author, trainer
- Joined
- 10/4/07
- Messages
- 10,469
- Points
- 648
Your post is not clear. Can you rephrase it.
Are you trying to optimize a multi-variable function?
Are you trying to optimize a multi-variable function?
- Joined
- 4/6/15
- Messages
- 24
- Points
- 11
Yes,exactly/I already know how many parameters it have . I am entering range and step. Range [0..10] Step 2. Which else algorithms for multi-variable function optimization do y know?
Daniel Duffy
C++ author, trainer
- Joined
- 10/4/07
- Messages
- 10,469
- Points
- 648
What is your function precisely and is it constrained etc.
- Joined
- 4/6/15
- Messages
- 24
- Points
- 11
It is time series analysing (think like black box with parameters)
There are strategy backtesting for example sma(6) crossover sma (100)
5 and 100 is my parameters A and B, I can (optional give them range and step for better optimising algorithm performaance) There are geentic algorithms for this task, particle swarm ETC
Which algorithms do you know for this task&What y can recommend&Any ideas?
Duffy i saw your profile ,you know a lot about optimisation
There are strategy backtesting for example sma(6) crossover sma (100)
5 and 100 is my parameters A and B, I can (optional give them range and step for better optimising algorithm performaance) There are geentic algorithms for this task, particle swarm ETC
Which algorithms do you know for this task&What y can recommend&Any ideas?
Duffy i saw your profile ,you know a lot about optimisation
- Joined
- 4/6/15
- Messages
- 24
- Points
- 11
http://stackoverflow.com/questions/1211201/free-optimization-library-in-c-sharp
What do y think about thislist
What i should use for multivariable function optimisation?
- Numerical provides a variety of algorithms including:
- Chromosome Manager
- Genetic Optimizer
- Hill Climbing Optimizer
- Maximizing Point
- Maximizing PointFactoy
- Maximizing Vector
- Minimizing Point
- Minimizing Point Factory
- Minimizing Vector
- Multi Variable General Optimizer
- Multi Variable Optimizer
- One Variable Function Optimizer
- Optimizing Bracket Finder
- Optimizing Point
- Optimizing Point Factory
- Optimizing Vector
- Simplex Optimizer
- Vector Chromosome Manager
- Vector Genetic Optimizer
- Vector Projected Function
Daniel Duffy
C++ author, trainer
- Joined
- 10/4/07
- Messages
- 10,469
- Points
- 648
Start here.
Duffy i saw your profile ,you know a lot about optimization
A little bit, not exactly my core skill set.
Last edited:
- Joined
- 4/6/15
- Messages
- 24
- Points
- 11
What does it mean Gradinet function?Can i use it for my task?
Which optimisation algorithms have boundaries (X (0...100) etc and (maybe steps)
Daniel Duffy
C++ author, trainer
- Joined
- 10/4/07
- Messages
- 10,469
- Points
- 648
Last edited:
- Joined
- 4/6/15
- Messages
- 24
- Points
- 11
L-BFGS and CG
Unconstrained optimization. Function gradient is needed.
Levenberg-Marquardt algorithm
Unconstrained or boundary constrained optimization.
Bound and linear equality/inequality constrained optimization
Linearly constrained optimization.
Constrained quadratic programming with box/linear constraints
ALL theese methods require gradints and looks like they are not for me
Or i am wrong&How i can write gradient?
You can help me if just say Which method i shoud choose? maybe you know one or two good?
- Joined
- 8/8/10
- Messages
- 343
- Points
- 53
>There are strategy backtesting for example sma(6) crossover sma (100)
If it is a real-life example, just try a brute force: a short SMA usually varies from 3 to 21 days (common value is 14) and the long SMA from 50 to 200 days (the latter is most common).
Totally you have about 3000 combinations, which is a piece of cake for a brute force search.
P.S.
I analyzed this problem and come to conclusion that:
1) MA Crossover beats Buy&Hold in case of German DAX but not in case of DJ30.
2) Optimization of MA-parameters does not bring anything significant, compared to SMA(14)+SMA(200)
If it is a real-life example, just try a brute force: a short SMA usually varies from 3 to 21 days (common value is 14) and the long SMA from 50 to 200 days (the latter is most common).
Totally you have about 3000 combinations, which is a piece of cake for a brute force search.
P.S.
I analyzed this problem and come to conclusion that:
1) MA Crossover beats Buy&Hold in case of German DAX but not in case of DJ30.
2) Optimization of MA-parameters does not bring anything significant, compared to SMA(14)+SMA(200)
- Joined
- 4/6/15
- Messages
- 24
- Points
- 11
> just try a brute force
Yes i am using brute force/But if i have 3- parametrs it is too long so i need another method. What can you reccomend?
- Joined
- 8/8/10
- Messages
- 343
- Points
- 53
Well, brute force will also cope with 3 parameters, esp. if they are integers (like number of days in SMA). If you know CUDA/GPU you can accelerate your code upto 1000 times, have a look at my paper(links to code is available in footnotes on page 9): http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2259133
If you don't know CUDA, there some compilers that can do the job for you, essentially you just need to add #pragma acc parallel before your loop: http://on-demand.gputechconf.com/gtc-express/2011/presentations/OpenACC_1_0_intro_jan2012.pdf
(Disclaimer, I didn't try OpenACC myself, when I need the power of GPU I usually program in CUDA directly).
Daniel Duffy
C++ author, trainer
- Joined
- 10/4/07
- Messages
- 10,469
- Points
- 648
This is a weird and confused thread that has nothing to do with the title
"Monte carlo simulation optimisation framework C#"
Feels more like "feed me".
"Monte carlo simulation optimisation framework C#"
Feels more like "feed me".
- Joined
- 4/6/15
- Messages
- 24
- Points
- 11
YES, monte carlo i just one of many oways to optimize function
I thought my question is very simple
But now i understood that i need not just simple monte carlo.
I am looking for multi level monte carlo
https://people.maths.ox.ac.uk/gilesm/mlmc.html
Daniel,maybe you know something about it? or know simillar algorithms?
https://people.maths.ox.ac.uk/gilesm/talks/mcqmc12_giles.pdf
I thought my question is very simple
But now i understood that i need not just simple monte carlo.
I am looking for multi level monte carlo
https://people.maths.ox.ac.uk/gilesm/mlmc.html
Daniel,maybe you know something about it? or know simillar algorithms?
https://people.maths.ox.ac.uk/gilesm/talks/mcqmc12_giles.pdf
- Joined
- 4/6/15
- Messages
- 24
- Points
- 11
Daniel i red your interview in moneyscience.Very intresting
I found on their site information about multi-level Monte-carlo, dynamic mc
Did you tried it?
www.moneyscience.com/mod/file/download.php?file_guid=545543
- Joined
- 4/6/15
- Messages
- 24
- Points
- 11
Daniel i found that you are author of Monte Carlo Frameworks in C++
by Daniel J. Duffy (Author), Joerg Kienitz (Author)
Can you reccommend something similaar framework in c# or c++.net?
Daniel Duffy
C++ author, trainer
- Joined
- 10/4/07
- Messages
- 10,469
- Points
- 648
I am an external advisor at unis.Here are theses from 2014 (See #1326297 => MLMC)
http://www.datasimfinancial.com/forum/viewtopic.php?f=198&t=1198
