saving the state of the system post-simulation

[Ayse]

Hi guys,

Is there any function in Dedalus which can save the state of the system after a simulation ends?  I am thinking of something compatible with the load_state function, which saves distributed field and writes down all the information about the state of the system.

For to clarify more, something as adding a file_handler and setting it to the last iteration does not help me since I need to save the state after another function ask the state to be saved. So this other function may skip many full simulations and decide to save after a certain condition is met.

Thank you in advance,

Ayse

[Evan H. Anders]

Hi Ayse,

The trick that we use for this is to add a checkpointing file handler at the end of the simulation run, then force it to take *one* more timestep to trigger the file handler to save the state.  We use a Checkpointing wrapper on the file handler class (attached, largely coded by Jeff Oishi a year+ ago), import the Checkpoint class from this file, and then add some lines like this at the end of the simulation:

                final_checkpoint = Checkpoint(data_dir, checkpoint_name='final_checkpoint')

                final_checkpoint.set_checkpoint(solver, wall_dt=1, mode="overwrite") # can also use mode='append' if you want to store multiple

                solver.step(dt)

This works for us, and is how we restart our simulations.  If there's a more elegant way, I'd love to know about it too!

Best,

Evan

--
You received this message because you are subscribed to the Google Groups "Dedalus Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dedalus-users+unsubscribe@googlegroups.com.
To post to this group, send email to dedalu...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/dedalus-users/6729d70b-287c-4911-bd7b-c1cd4e5ad8c9%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ayse]

Hi Evan,

Thank you for your quick reply. Is it possible for you to send me a very simple working example that shows how to use this check-pointing to save and restart?

Best,
Ayse

On Friday, January 12, 2018 at 2:47:37 PM UTC+1, Evan H. Anders wrote:

Hi Ayse,

The trick that we use for this is to add a checkpointing file handler at the end of the simulation run, then force it to take *one* more timestep to trigger the file handler to save the state.  We use a Checkpointing wrapper on the file handler class (attached, largely coded by Jeff Oishi a year+ ago), import the Checkpoint class from this file, and then add some lines like this at the end of the simulation:

                final_checkpoint = Checkpoint(data_dir, checkpoint_name='final_checkpoint')

                final_checkpoint.set_checkpoint(solver, wall_dt=1, mode="overwrite") # can also use mode='append' if you want to store multiple

                solver.step(dt)

This works for us, and is how we restart our simulations.  If there's a more elegant way, I'd love to know about it too!

Best,

Evan

On Fri, Jan 12, 2018 at 6:18 AM, Ayse <ays...@gmail.com> wrote:

Hi guys,

Is there any function in Dedalus which can save the state of the system after a simulation ends?  I am thinking of something compatible with the load_state function, which saves distributed field and writes down all the information about the state of the system.

For to clarify more, something as adding a file_handler and setting it to the last iteration does not help me since I need to save the state after another function ask the state to be saved. So this other function may skip many full simulations and decide to save after a certain condition is met.

Thank you in advance,

Ayse

--
You received this message because you are subscribed to the Google Groups "Dedalus Users" group.

To unsubscribe from this group and stop receiving emails from it, send an email to dedalus-user...@googlegroups.com.

[Evan H. Anders]

Hey Ayse,

Hmmm. I don't have a very simple working example on hand, unfortunately.  I usually use this in fully compressible convection sims (our repo of this just went public yesterday, https://bitbucket.org/exoweather/polytrope/overview, but the code is a little bit of a mess and hard ish to understand).  I think that's the most public working example of this we have.  The driver file is FC_poly.py (https://bitbucket.org/exoweather/polytrope/src/488c455091f9d25229af52d2533d0da27eb0421c/FC_poly.py?at=default&fileviewer=file-view-default), and lines 160-165 are where we restart from checkpoint (or use initial conditions), and then lines 286-299 are the ones where we set up the final checkpointing.

So you could run the script like (I can do this quickly on my local processor, doesn't get past transient, but it runs):

python3 FC_poly.py --Rayleigh=1e2 --nz=32 --nx=64 --run_time_buoy=20

and then restart from where you left off afterwards by doing:

python3 FC_poly.py --Rayleigh=1e2 --nz=32 --nx=64 --run_time_buoy=20 --restart=FC_poly_2D_nrhocz3_Ra1e2_Pr1_eps1e-4_a4/final_checkpoint/final_checkpoint_s1.h5

and it'll pick up from where it left off.  Basically if you ctrl+f "checkpoint" on the FC_poly.py file, you'll find all the lines used to make this happen (there aren't many), but there are a lot of extra cruft lines in that file for other options.

Sorry that's not the cleanest answer, let me know if I can clarify!

Evan

To unsubscribe from this group and stop receiving emails from it, send an email to dedalus-users+unsubscribe@googlegroups.com.

To post to this group, send email to dedalu...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/dedalus-users/73c7e48f-9098-4b63-ba01-9514daff79f1%40googlegroups.com.

[Ayse]

Hi again Evan,

Thank you. I will play with it, if I stuck I will write again.

Have a nice day

Ayse

[Jeffrey S. Oishi]

Hi Ayse and all,

I just wanted to follow up on this. Here is a simple solution to do this that does not require any extra code. 

If you have a file handler for your system, say one called checkpoint, you can simply do 

# for example

checkpoint = solver.add_file_handler('checkpoint', etc)

checkpoint.add_system(solver.state)

while solver.ok:

    <main loop goes here>

# write last checkpoint

end_world_time = solver.get_world_time()

end_wall_time = end_world_time - solver.start_time

solver.evaluator.evaluate_handlers([checkpoint,], timestep = dt, sim_time = solver.sim_time, world_time=end_world_time, wall_time=end_wall_time, iteration=solver.iteration)

I've assumed your solver is called solver. Please note that the filehandler must be in a list when passing it to evalue_handlers. 

Please let me know if you have any questions.

Jeff

To view this discussion on the web visit https://groups.google.com/d/msgid/dedalus-users/554c4eeb-61e3-4398-95ed-6d7b3f7443da%40googlegroups.com.

[Ayse Yesil]

Hi Jeff,

Thank you for the follow up. I will have a look and come back to this post.

Cheers

[Louis-Alexandre Couston]

Hi Jeff,

Can we still do this trick in d3 to save data of the final step as a checkpoint file?

It's returning an error about solver not having the get_world_time() argument.

Thanks,

Louis

[Jeffrey S. Oishi]

Hi Louis,

Yes, and it's much simpler now:

solver.evaluate_handlers_now(dt)

where dt is the current timestep.

Jeff

> To view this discussion on the web visit https://groups.google.com/d/msgid/dedalus-users/3e0cf854-3185-4b5b-b493-b0f056a1c9e7n%40googlegroups.com.

[Louis-Alexandre Couston]

oh that's great! Thanks :)

You received this message because you are subscribed to a topic in the Google Groups "Dedalus Users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/dedalus-users/fUj_q91F9bo/unsubscribe.
To unsubscribe from this group and all its topics, send an email to dedalus-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/dedalus-users/CAAOZTN-hyZ0kTWBXjqcVCW_%2B5CEU_MmUQEvY%2BWzC%2BRE64EiC9Q%40mail.gmail.com.

[Gustavo Estay]

Hello,

Just for future reference, it seems that from May 17, 2023, solver.evaluate_handlers_now(dt) is deprecated and is equivalent to evaluate_handlers(handlers=handlers, dt=dt).

Best,

Gustavo