Allow resuming simulation by regularly saving to file #5
Conversation
pablolh
force-pushed
the
plh/autosave
branch
4 times, most recently
from
3646427 to
a5c6145
Compare
pablolh
changed the title
|
It is designed to work with all sorts of noise. I tried it in a bunch of cases, notably the noisy benchmarks were tested for same plots after the previous MR. |
In this case, then where is stored the number of runs of the monte carlo and in which run the code is actually running? or how is the code managing this situation? |
Having several runs for Monte Carlo in one go was something we supported in emu-tn, but here it assumes a single run. If you want to have several runs, you need to write your own code to do it. This code allows you to checkpoint an individual run. You could build a checkpointing system for many runs out of this. |
|
fixed the logging and default autosave time interval |
- Adds autosave_prefix and autosave_dt config fields
By default:
emu_mps_save_<uuid>.dat
every 10 minute
- Accidentally fix the various config entries and
carry the whole MPSConfig along. It was part of my
research for optimal implementation and takes too
much time to separate as a independent MR.
- Some fixes to the unpickled impl are necessary,
like autosave_file (to save to same existing file)
and the logger which gets messed up by the pickling.
- Fix make_H modifying its interaction_matrix argument.
That was also part of the process but in fact now
unrelated.
---> How to resume a simulation from a file
import emu_mps
results = emu_mps.MPSBackend().resume("emu_mps_save_xxx.dat")
By default:
emu_mps_save_<uuid>.datevery 10 minute
carry the whole MPSConfig along. It was part of my
research for optimal implementation and takes too
much time to separate as a independent MR.
like autosave_file (to save to same existing file)
and the logger which gets messed up by the pickling.
That was also part of the process but in fact now
unrelated.
---> How to resume a simulation from a file