log simulations and save intermediate numpys

[brookehus]

Development:

• [x] Implement feature / fix bug
• [x] Add documentation

• [x] Add tests

  Checks:

• [x] Run nosetests
• [x] Check pep8 compliance

SaveTheNumpys

This PR:

• Removes the verbose argument of the simulation class
• Adds four arguments: save_npys, log, log_type, and filename
• save_npys and log default to None. Otherwise, they can be a number. If they are an integer, then that number of saves/logs are performed at regular intervals. If they are between 0 and 1, then saves/logs are performed at every % according to that argument.
• log_type can be 'print' or 'write'. if print, it prints. if write, it writes to file
• filename specifies the base file name, to which _log.txt will be appended for logs, and _0001.npy (etc.) will be appended for numpys. Overwriting is not allowed.

TODO:

• [x] Implement print logging
• [x] Implement file logging
• [x] Implement numpy saving
• [ ] Decide how to handle nan checking (and implement) (moved to a new issue)
• [x] Documentation
• [x] Testing?
• [x] Fix example notebooks that call verbose

@nec4 if you have comments on the overall strategy, please let me know! And feel free to look over the code for the logging part.

[nec4]

Looks great - will take a look!

[brookehus]

ok, so there are 2 ways that save_npys and log are supposed to work: you either: (1) give them an integer 1 or greater, in which case they save that number of files/log that number of steps (or greater, in the case of non-even division), or (2) give them a float between 0 and 1, in which case they save files/log at that % completion.

(2) is a little wonky right now. i've written some tests that pass but i'm not totally sure why. anyway, give this a look, and i'll hopefully be able to approach it with a clearer mind tomorrow!

[brookehus]

the last four tests - i.e. the ones with _float at the end - may be totally meaningless

[nec4]

Great work! I'll take another look!

[nec4]

There is a lot here - great work! I have a general question - what is the interplay between _npy_interval and save_interval and log_interval (if any)? self.simulated_traj and the other attribute observables are updated in _save_timepoint() but they are only written to file by the function _save_numpy(), which accesses self.simulated_traj and the other attributes when (t+1) % _npy_interval == 0. Will there ever be an issue from having class attribute observables updated by save_interval but the file saved attributes updated by _npy_interval? I may be overthinking this.

[brookehus]

I think your point is good, and to save everyone including myself right now a lot of trouble, I think I will restrict the numpy interval to be only possible at saving interval points

[brookehus]

@nec4 can you resolve solved conversations?

I need to look at the example notebooks and fix any verbose flags, and then we talked about getting the nan checker to return the iteration at which a nan showed up.

Anything else you want to see? Can you look over the latest code?

Update: fixed example notebooks in f51f82e

[brookehus]

Ok @nec4, in addition to standard code review, can you:

• Resolve all pending conversations
• Go through the nosetests and ensure example notebooks run
• Using the tests as inspiration, actually play around and make sure things work like you expect them to
• Determine if you think any further tests are needed

[nec4]

Sounds good - will take a look now!

[nec4]

I think this looks more clean than before - great work! Aside from a small typo I can't think of anything else to add/change.

[brookehus]

you can decide whether to merge or wait to play around!

[nec4]

LGTM!