search

lisa-gm / INLA_DIST

BSD 3-Clause "New" or "Revised" License
8 stars 0 forks source link

inla3d #1

Open yy177399-yan opened 5 months ago

yy177399-yan commented 5 months ago

hello

I have some questions about inla, Can it include variables in it? such as: "result2 <- inla(som ~ -1+AMT+Aspect+Clay+DEM+MrVBF+NDVI+PLC+PRC+Slope+Soiltype+TRI+TWI+f(field, model = spde),family="gaussian", data = inla.stack.data(stack, spde = spde), control.predictor=list(A=inla.stack.A(stack), compute=TRUE), control.compute=list(cpo=FALSE,dic=TRUE), control.mode = list(theta=log(c(sd_0^-2,range_s_0,sigma_0))+0.5,restart=TRUE), control.inla = list(reordering = "metis"), keep=FALSE, verbose=TRUE)"; is it right??

lisa-gm commented 4 months ago

Hi,

Thanks for your interest in the repo! It's currently still under development and doesn't have the standard inla interface interface yet, sorry. What are you trying to model? And if you are interested we can keep you updated once this exists.

Best, Lisa

On Tue, Feb 6, 2024 at 9:51 AM yy177399-yan @.***> wrote:

hello

I have some questions about inla3d ( https://github.com/search?q=inla3dt&type=repositories), Can it include variables in it? such as: "result2 <- inla(som ~ -1+ AMT+Aspect+Clay+DEM+MrVBF+NDVI+PLC+PRC+Slope+Soiltype+TRI+TWI+f(field, model = spde),family="gaussian", data = inla.stack.data(stack, spde = spde), control.predictor=list(A=inla.stack.A(stack), compute=TRUE), control.compute=list(cpo=FALSE,dic=TRUE), control.mode = list(theta=log(c(sd_0^-2,range_s_0,sigma_0))+0.5,restart=TRUE), control.inla = list(reordering = "metis"), keep=FALSE, verbose=TRUE)"; is it right??

— Reply to this email directly, view it on GitHub https://github.com/lisa-gm/INLA_DIST/issues/1, or unsubscribe https://github.com/notifications/unsubscribe-auth/AK3SAV53UQEN7QGZZSOIPJLYSHVIJAVCNFSM6AAAAABC3SDDCOVHI2DSMVQWIX3LMV43ASLTON2WKOZSGEZDAMRWGAYDAOI . You are receiving this because you are subscribed to this thread.Message ID: @.***>

yy177399-yan commented 4 months ago 
    Dear Lisa,I hope this email finds you well. I am currently engaged in constructing large-scale soil carbon models using the Integrated Nested Laplace Approximation (INLA) framework with Stochastic Partial Differential Equation (SPDE) approach in R. Given the complexity and scale of these models, I have encountered significant computational challenges, including slow processing times and high demands on computer resources.In an effort to address these issues, I am exploring the possibility of leveraging Graphics Processing Unit (GPU) acceleration to improve computational efficiency. Before proceeding, I wanted to reach out to inquire about the following:Does the INLA package in R currently support GPU acceleration for the type of models I am working on? If so, could you provide guidance on how to enable and optimally utilize GPU support?Beyond GPU acceleration, do you have any additional recommendations or best practices for enhancing the computational performance of large-scale INLA-SPDE models in R? This might include suggestions on coding practices, model optimization techniques, or alternative computational resources.If GPU support is not available, are there plans to incorporate this feature in future releases of the INLA package in R, or are there other software tools within the R ecosystem that you would recommend for this purpose?The success of my research heavily relies on the ability to efficiently process and analyze large datasets. Any insights or advice you could provide would be invaluable and greatly appreciated.Thank you very much for your time and consideration. I look forward to your response and any recommendations you might have.Best regards, Yang

                    ***@***.***

---- Replied Message ----

     From 

        ***@***.***>

     Date 

    2/10/2024 18:20

     To 

        ***@***.***>

     Cc 

        ***@***.***>
        ,

        ***@***.***>

     Subject 

          Re: [lisa-gm/INLA_DIST] inla3d (Issue #1)

Hi,

Thanks for your interest in the repo! It's currently still under

development and doesn't have the standard inla interface interface yet,

sorry. What are you trying to model? And if you are interested we can keep

you updated once this exists.

Best,

Lisa

On Tue, Feb 6, 2024 at 9:51 AM yy177399-yan @.***>

wrote:

hello

I have some questions about inla3d (

https://github.com/search?q=inla3dt&type=repositories), Can it include

variables in it? such as: "result2 <- inla(som ~ -1+

AMT+Aspect+Clay+DEM+MrVBF+NDVI+PLC+PRC+Slope+Soiltype+TRI+TWI+f(field,

model = spde),family="gaussian",

data = inla.stack.data(stack, spde = spde),

control.predictor=list(A=inla.stack.A(stack), compute=TRUE),

control.compute=list(cpo=FALSE,dic=TRUE),

control.mode =

list(theta=log(c(sd_0^-2,range_s_0,sigma_0))+0.5,restart=TRUE),

control.inla = list(reordering = "metis"),

keep=FALSE, verbose=TRUE)";

is it right??

—

Reply to this email directly, view it on GitHub

https://github.com/lisa-gm/INLA_DIST/issues/1, or unsubscribe

https://github.com/notifications/unsubscribe-auth/AK3SAV53UQEN7QGZZSOIPJLYSHVIJAVCNFSM6AAAAABC3SDDCOVHI2DSMVQWIX3LMV43ASLTON2WKOZSGEZDAMRWGAYDAOI

.

You are receiving this because you are subscribed to this thread.Message

ID: @.***>

—Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: @.***>

lisa-gm commented 3 months ago

Hi!

Thanks for the detailed information.

R-INLA currently does not offer GPU support at the moment. It's non-trivial for this to actually lead to significant performance improvements as INLA heavily relies on sparse linear algebra operations. We are discussing to add this option though but for sure it won't be available in the near future.

If I read your inla() call from the previous e-mail correctly you have normally distributed data, a spatial SPDE model and then some linear covariates. Is that correct? If so, we could try to make this run with INLA_DIST (these types of models are supported) if you are happy to accept the lack of a nice user interface and some custom tuning ... This would probably require a bit of work though to make it run. Once it runs, I expect it to be significantly faster.

How large is your spatial domain (i.e. how many nodes are in your spatial mesh)?

Best, Lisa

On Wed, Mar 6, 2024 at 9:10 AM yy177399-yan @.***> wrote:

Dear Lisa,I hope this email finds you well. I am currently engaged in constructing large-scale soil carbon models using the Integrated Nested Laplace Approximation (INLA) framework with Stochastic Partial Differential Equation (SPDE) approach in R. Given the complexity and scale of these models, I have encountered significant computational challenges, including slow processing times and high demands on computer resources.In an effort to address these issues, I am exploring the possibility of leveraging Graphics Processing Unit (GPU) acceleration to improve computational efficiency. Before proceeding, I wanted to reach out to inquire about the following:Does the INLA package in R currently support GPU acceleration for the type of models I am working on? If so, could you provide guidance on how to enable and optimally utilize GPU support?Beyond GPU acceleration, do you have any additional recommendations or best practices for enhancing the computational performance of large-scale INLA-SPDE models in R? This might include suggestions on coding practices, model optimization techniques, or alternative computational resources.If GPU support is not available, are there plans to incorporate this feature in future releases of the INLA package in R, or are there other software tools within the R ecosystem that you would recommend for this purpose?The success of my research heavily relies on the ability to efficiently process and analyze large datasets. Any insights or advice you could provide would be invaluable and greatly appreciated.Thank you very much for your time and consideration. I look forward to your response and any recommendations you might have.Best regards, Yang

@.***

---- Replied Message ----

From

@.***>

Date

2/10/2024 18:20

To

@.***>

Cc

@.***> ,

@.***>

Subject

Re: [lisa-gm/INLA_DIST] inla3d (Issue #1)

Hi,

Thanks for your interest in the repo! It's currently still under

development and doesn't have the standard inla interface interface yet,

sorry. What are you trying to model? And if you are interested we can keep

you updated once this exists.

Best,

Lisa

On Tue, Feb 6, 2024 at 9:51 AM yy177399-yan @.***>

wrote:

hello

I have some questions about inla3d (

https://github.com/search?q=inla3dt&type=repositories), Can it include

variables in it? such as: "result2 <- inla(som ~ -1+

AMT+Aspect+Clay+DEM+MrVBF+NDVI+PLC+PRC+Slope+Soiltype+TRI+TWI+f(field,

model = spde),family="gaussian",

data = inla.stack.data(stack, spde = spde),

control.predictor=list(A=inla.stack.A(stack), compute=TRUE),

control.compute=list(cpo=FALSE,dic=TRUE),

control.mode =

list(theta=log(c(sd_0^-2,range_s_0,sigma_0))+0.5,restart=TRUE),

control.inla = list(reordering = "metis"),

keep=FALSE, verbose=TRUE)";

is it right??

—

Reply to this email directly, view it on GitHub

https://github.com/lisa-gm/INLA_DIST/issues/1, or unsubscribe

< https://github.com/notifications/unsubscribe-auth/AK3SAV53UQEN7QGZZSOIPJLYSHVIJAVCNFSM6AAAAABC3SDDCOVHI2DSMVQWIX3LMV43ASLTON2WKOZSGEZDAMRWGAYDAOI

.

You are receiving this because you are subscribed to this thread.Message

ID: @.***>

—Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: @.***>

— Reply to this email directly, view it on GitHub https://github.com/lisa-gm/INLA_DIST/issues/1#issuecomment-1980299882, or unsubscribe https://github.com/notifications/unsubscribe-auth/AK3SAV75XVNZXCOB55DVWVTYW3FOVAVCNFSM6AAAAABC3SDDCOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSOBQGI4TSOBYGI . You are receiving this because you commented.Message ID: @.***>