[julia-users] Re: Considering posting SDE solver package

2015-01-19 Thread mschauer
Hi, Micah, Alex
Currently I am stalled down in development of the package because of the 
lack of FixedSizedArrays support in the language itself. This is tagged 
0.4, https://github.com/JuliaLang/julia/pull/7568 and is important for very 
fast small-dimensional many data problems.
I planned to wait for that and then refactor the package. On the other 
hand, if you are looking at a high dimensional processes, this is maybe not 
so important and fast ArrayViews are sufficient. Anyway, we should be able 
to bridge our materials, if we want to. 

Best regards,
Moritz


On Monday, January 19, 2015 at 6:40:49 AM UTC+1, Micah Corah wrote:
>
> Alex,
>
> Thanks for pointing out SDE.jl. If I continue, a name like 
> StochasticDifferentialEquations.jl or Stochastic.jl would do. I should have 
> found that earlier. The author clearly knows much more about the material 
> than I do and has spent a lot of time developing the package. SDE.jl also 
> seems to have a fairly broad set of functionality more tailored to the 
> scientific community. However, it looks largely abandoned at the moment. 
> The documentation looks pretty good, and it provides a simiilar 
> implementation of Euler Maruyama.
>
> Avik,
>
> This really just comes down to a matter of preference. I personally prefer 
> a more self-contained solution such as used with most ODE solvers, but that 
> may be just a matter of naivete. Mostly, it just seems thin to me. I had 
> considered wrapping Ito at the time, but it didn't seem that I would gain 
> much by doing so. In retrospect, I did not understand Julia as well at the 
> time, and the usage of StochasticProcess made less sense to me at the time.
>
>
> My personal use cases will tend toward simulation of control systems and 
> multi-robot systems. These use cases are likely simpler with less strict 
> requirements than other applications. In my case speed is the biggest issue 
> as I have been working with simulations having tens of thousands of 
> variables but otherwise just having a linear system with noise. Thus, 
> although I have made an effort to get to know the theory better (and will 
> continue to if I decide to continue to pursue this project) my 
> understanding of SDEs is probably still fairly simplistic.
>
> It looks to me, like much of what I would be able to provide has already 
> been done. The need that I thought existed might just be a matter of lack 
> of publicity. If anyone still sees a need for a different package for any 
> of the reasons I mentioned (or otherwise for that matter) I will consider 
> continued development. Otherwise, I will either leave as-is or continue 
> development with a focus on personal use.
>
> Thanks for all the help.
>
> Micah
>
> On Sunday, January 18, 2015 at 5:57:59 PM UTC-7, Avik Sengupta wrote:
>>
>> Hi Micah, 
>>
>> I intend Ito to only contain specific financial products/algos. My 
>> thinking has been that all core math should be in external packages. Having 
>> said that, I'd be happy for any feedback on Ito itself. Please feel free to 
>> open issues with any comments on taste or correctness. 
>>
>> Regards
>> -
>> Avik
>>
>> On Sunday, 18 January 2015 06:21:41 UTC, Micah Corah wrote:
>>>
>>> I have been working on a project involving large systems of stochastic 
>>> differential equations. I considered using Ito.jl (to my knowledge the only 
>>> package with features for SDEs), but it was not quite to my tastes and did 
>>> not seem to offer much. In response, I wrote my own solver based on the 
>>> tutorial "An Algorithmic Introduction to Numerical Simulation of Equations" 
>>> by Desmond Higham (Euler-Maruyama) and have spent some time optimizing it. 
>>> Would this be something that people would be interested in using?
>>>
>>>
>>> There are some small changes that I would have to make for theoretical 
>>> correctness, but otherwise it looks fairly nice to me. If I do turn this 
>>> into its own package I will fix any of these latent issues and also work on 
>>> adding support for features such as different algorithms and distributions. 
>>> Then we would have Ito.jl for quantitative finance and SDE.jl (or whatever 
>>> I call it) for controls, dynamical systems, etc.
>>>
>>>
>>> Currently this code is on github with the code for a personal project I 
>>> am working on at https://github.com/mcorah/MultiQuadLift
>>>
>>

[julia-users] Re: Considering posting SDE solver package

2015-01-18 Thread Micah Corah
Alex,

Thanks for pointing out SDE.jl. If I continue, a name like 
StochasticDifferentialEquations.jl or Stochastic.jl would do. I should have 
found that earlier. The author clearly knows much more about the material 
than I do and has spent a lot of time developing the package. SDE.jl also 
seems to have a fairly broad set of functionality more tailored to the 
scientific community. However, it looks largely abandoned at the moment. 
The documentation looks pretty good, and it provides a simiilar 
implementation of Euler Maruyama.

Avik,

This really just comes down to a matter of preference. I personally prefer 
a more self-contained solution such as used with most ODE solvers, but that 
may be just a matter of naivete. Mostly, it just seems thin to me. I had 
considered wrapping Ito at the time, but it didn't seem that I would gain 
much by doing so. In retrospect, I did not understand Julia as well at the 
time, and the usage of StochasticProcess made less sense to me at the time.


My personal use cases will tend toward simulation of control systems and 
multi-robot systems. These use cases are likely simpler with less strict 
requirements than other applications. In my case speed is the biggest issue 
as I have been working with simulations having tens of thousands of 
variables but otherwise just having a linear system with noise. Thus, 
although I have made an effort to get to know the theory better (and will 
continue to if I decide to continue to pursue this project) my 
understanding of SDEs is probably still fairly simplistic.

It looks to me, like much of what I would be able to provide has already 
been done. The need that I thought existed might just be a matter of lack 
of publicity. If anyone still sees a need for a different package for any 
of the reasons I mentioned (or otherwise for that matter) I will consider 
continued development. Otherwise, I will either leave as-is or continue 
development with a focus on personal use.

Thanks for all the help.

Micah

On Sunday, January 18, 2015 at 5:57:59 PM UTC-7, Avik Sengupta wrote:
>
> Hi Micah, 
>
> I intend Ito to only contain specific financial products/algos. My 
> thinking has been that all core math should be in external packages. Having 
> said that, I'd be happy for any feedback on Ito itself. Please feel free to 
> open issues with any comments on taste or correctness. 
>
> Regards
> -
> Avik
>
> On Sunday, 18 January 2015 06:21:41 UTC, Micah Corah wrote:
>>
>> I have been working on a project involving large systems of stochastic 
>> differential equations. I considered using Ito.jl (to my knowledge the only 
>> package with features for SDEs), but it was not quite to my tastes and did 
>> not seem to offer much. In response, I wrote my own solver based on the 
>> tutorial "An Algorithmic Introduction to Numerical Simulation of Equations" 
>> by Desmond Higham (Euler-Maruyama) and have spent some time optimizing it. 
>> Would this be something that people would be interested in using?
>>
>>
>> There are some small changes that I would have to make for theoretical 
>> correctness, but otherwise it looks fairly nice to me. If I do turn this 
>> into its own package I will fix any of these latent issues and also work on 
>> adding support for features such as different algorithms and distributions. 
>> Then we would have Ito.jl for quantitative finance and SDE.jl (or whatever 
>> I call it) for controls, dynamical systems, etc.
>>
>>
>> Currently this code is on github with the code for a personal project I 
>> am working on at https://github.com/mcorah/MultiQuadLift
>>
>

[julia-users] Re: Considering posting SDE solver package

2015-01-18 Thread Avik Sengupta
Hi Micah, 

I intend Ito to only contain specific financial products/algos. My thinking 
has been that all core math should be in external packages. Having said 
that, I'd be happy for any feedback on Ito itself. Please feel free to open 
issues with any comments on taste or correctness. 

Regards
-
Avik

On Sunday, 18 January 2015 06:21:41 UTC, Micah Corah wrote:
>
> I have been working on a project involving large systems of stochastic 
> differential equations. I considered using Ito.jl (to my knowledge the only 
> package with features for SDEs), but it was not quite to my tastes and did 
> not seem to offer much. In response, I wrote my own solver based on the 
> tutorial "An Algorithmic Introduction to Numerical Simulation of Equations" 
> by Desmond Higham (Euler-Maruyama) and have spent some time optimizing it. 
> Would this be something that people would be interested in using?
>
>
> There are some small changes that I would have to make for theoretical 
> correctness, but otherwise it looks fairly nice to me. If I do turn this 
> into its own package I will fix any of these latent issues and also work on 
> adding support for features such as different algorithms and distributions. 
> Then we would have Ito.jl for quantitative finance and SDE.jl (or whatever 
> I call it) for controls, dynamical systems, etc.
>
>
> Currently this code is on github with the code for a personal project I am 
> working on at https://github.com/mcorah/MultiQuadLift
>


[julia-users] Re: Considering posting SDE solver package

2015-01-18 Thread Alex
Hi Micah,

This sounds great! I would be very interested in a dedicated SDE package 
and I guess others would be as well. At some point I wrote a solver based 
on the Heun method, but it needs some cleanup before I dare to post it.

AFAIK the name SDE.jl  is already taken 
...

Best,

Alex. 

On Sunday, 18 January 2015 07:21:41 UTC+1, Micah Corah wrote:
>
> I have been working on a project involving large systems of stochastic 
> differential equations. I considered using Ito.jl (to my knowledge the only 
> package with features for SDEs), but it was not quite to my tastes and did 
> not seem to offer much. In response, I wrote my own solver based on the 
> tutorial "An Algorithmic Introduction to Numerical Simulation of Equations" 
> by Desmond Higham (Euler-Maruyama) and have spent some time optimizing it. 
> Would this be something that people would be interested in using?
>
>
> There are some small changes that I would have to make for theoretical 
> correctness, but otherwise it looks fairly nice to me. If I do turn this 
> into its own package I will fix any of these latent issues and also work on 
> adding support for features such as different algorithms and distributions. 
> Then we would have Ito.jl for quantitative finance and SDE.jl (or whatever 
> I call it) for controls, dynamical systems, etc.
>
>
> Currently this code is on github with the code for a personal project I am 
> working on at https://github.com/mcorah/MultiQuadLift
>