I have noticed, that from some commit of osmps, hash-titled bin files started to appear in outputs directory. When doing calculations on external computer array, and post analysis on home computer, downloading these large binaries takes a lot of time, thus I started to think are they neccessary.
When doing conventional post analysis:
Defining operators, Hamiltonian, convergence parameters etc.;
calling: MainFiles = mps.WriteFiles(parameters, Operators, H, PostProcess=PostProcess);
retrieving Outputs;
Those binaries seem important, but when I remove call to mps.WriteFiles and do straigtforward reading of Observables, binaries seem obsolete.
Thus, it LOOKS LIKE binaries are a waste of time for post analysis, but trying to prevent corrupted or misread results I am asking you, dear developers, are there any instances where I must keep them?
Thanks in advance,
MrModern
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
the quick answer (before you download the next data package) is: For PostAnalysis, you don't need to download them. I usually delete them myself. But keep the [...]mapping.dat which is new as well.
The long answer and the idea behind: The hash files are - at the moment - the way to save states. I have chosen to hash the parameters in order to being able to detect if real time evolutions access the same initial state. Then the ground state is not recalculated, even if you run it with two different python file (as long as the output folder is the same). We aim as well for some measurement routine, where you could load those states for additional measurements.
In the case of the real time evolution, the mps of the latest measurement is saved. This becomes handy if the final point of your time evolution should be the starting point of a second time evolution. Although it is possible to continue your time evolution at the last point in case it was killed on a cluster.
If anybody needs a more detailed description of how to do those things, I could send or post a sketch of the documentation of this part.
I would appreciate your feedback, if you would prefer some kind of flag that this information should/should not be stored.
Kind regards,
Daniel
P.S. In context with the hashed states, some have already notices the new exception raised with "Cannot rerun simulation with different params". The issue behind here. You setup a simulation with system size L=100 (example only, works for any other parameter). Once the simulation run, other simulation with the same hash noticed that they can use this result (state and measurements!). If you change now to L=101, the hash changes and you cannot store it under the same name in order to keep the mapping correct for the other simulations.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks for the quick answer. Up until now, I have not used real time evolution, and I do not
intend to use it near future, thus it would be helpful for me to have a flag in parameters to shut off storage of these binaries.
MrModern
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Dear all,
I have noticed, that from some commit of osmps, hash-titled bin files started to appear in outputs directory. When doing calculations on external computer array, and post analysis on home computer, downloading these large binaries takes a lot of time, thus I started to think are they neccessary.
When doing conventional post analysis:
Defining operators, Hamiltonian, convergence parameters etc.;
calling: MainFiles = mps.WriteFiles(parameters, Operators, H, PostProcess=PostProcess);
retrieving Outputs;
Those binaries seem important, but when I remove call to mps.WriteFiles and do straigtforward reading of Observables, binaries seem obsolete.
Thus, it LOOKS LIKE binaries are a waste of time for post analysis, but trying to prevent corrupted or misread results I am asking you, dear developers, are there any instances where I must keep them?
Thanks in advance,
MrModern
Hallo MrModern,
the quick answer (before you download the next data package) is: For PostAnalysis, you don't need to download them. I usually delete them myself. But keep the [...]mapping.dat which is new as well.
The long answer and the idea behind: The hash files are - at the moment - the way to save states. I have chosen to hash the parameters in order to being able to detect if real time evolutions access the same initial state. Then the ground state is not recalculated, even if you run it with two different python file (as long as the output folder is the same). We aim as well for some measurement routine, where you could load those states for additional measurements.
In the case of the real time evolution, the mps of the latest measurement is saved. This becomes handy if the final point of your time evolution should be the starting point of a second time evolution. Although it is possible to continue your time evolution at the last point in case it was killed on a cluster.
If anybody needs a more detailed description of how to do those things, I could send or post a sketch of the documentation of this part.
I would appreciate your feedback, if you would prefer some kind of flag that this information should/should not be stored.
Kind regards,
Daniel
P.S. In context with the hashed states, some have already notices the new exception raised with "Cannot rerun simulation with different params". The issue behind here. You setup a simulation with system size L=100 (example only, works for any other parameter). Once the simulation run, other simulation with the same hash noticed that they can use this result (state and measurements!). If you change now to L=101, the hash changes and you cannot store it under the same name in order to keep the mapping correct for the other simulations.
Hi Daniel,
Thanks for the quick answer. Up until now, I have not used real time evolution, and I do not
intend to use it near future, thus it would be helpful for me to have a flag in parameters to shut off storage of these binaries.
MrModern