uw-comphys
Resume of Simulation From Last Saved File
Added ability to resume the simulation from the last saved file. This implementation manually requires the user to specify the name of the last time-step.
A version that works automatically will require much more changes to the way that the initial condition is handled.
I've split the mutable default arguments into a seperate PR (#46). All that's remaining now are the changes needed for resuming from a previous result.
I think there are a few things that need to be done before this functionality can be added:
- discussion with others regarding how this functionality should be configured and perform.
- addition of documentation of this functionality, preferably in a stationary AND transient example.
- changing the code in base_model.py to resume for only a stationary solve, transient solves are to be moved into a new derived solver class in the near future.
I think there are a few things that need to be done before this functionality can be added:
- discussion with others regarding how this functionality should be configured and perform.
- addition of documentation of this functionality, preferably in a stationary AND transient example.
- changing the code in base_model.py to resume for only a stationary solve, transient solves are to be moved into a new derived solver class in the near future.
I agree that the documentation is missing and I will add it. But I am unclear of what you mean for the other two points. We currently do not export intermediary iterations for a stationary solves. And even if we did, those can be loaded in simply as a the initial condition for a stationary solve, the stationary solves don't require anything to be done to them.
This is added for the transient simulations since resuming from a previous solve requires both changing the initial condition and the initial time value. That initial time is the only thing being added by this (that and a sanity check). The different initial time is needed for time varying BCs, forcing functions, etc. as well as for producing a correct time-series out of the results. Stationaryu solves don't have these considerations.
Do we really need to have this conversation on Github...your office is, literally, down the hall.