Major iteration counting

[ewu63]

Addresses #21 by:

1. calling the SNOPT routine snkerc instead of snoptc
2. writing a custom snstop code in Python to replace the one supplied by SNOPT, and called every major iteration
3. wrapping the snstop Python code with f2py and passing it into snkerc since snoptc does not allow passing a custom snstop function handle
4. modifying the db file structure such that the flag isMajor is set to true when the evaluation is at a major iteration
5. updating optview accordingly to filter iterations based on this flag

[ewu63]

To test, run test_hs015.py under /test, and examine the resulting Hist.hst file both by loading it in Python, and by opening it with optview. The flag isMajor should be present and true for major iterations which match the SNOPT print out file. Note that this implementation is able to determine which line search iterations is the actual major iteration, for example callCounter 11 is not a major iteration despite being a funcs call, and this is evident in the SNOPT print out, where nCon is incremented by 2 in major iteration 4.

[johnjasa]

@nwu63 what's the status on this? Should I test this out now?

[ewu63]

Nobody reviewed/tested the code so never merged. I did bug @eirikurj about it but I know he's been very busy. Would be great if someone can check to make sure it works as expected and then pull it in.

[johnjasa]

Need to change OptView logic to work well with other optimizers. Running with SLSQP then selecting Major Iterations on OptView produces a single point, the first iteration. Try with p hs015.py --opt=SLSQP --storeHistory=1 then running OptView on it.

Otherwise looks good on my end, with all tests passing and OptView looking good for SNOPT!

[johnjasa]

More changes requested:
With the proposed changes, you can't read older history files because they fail on a KeyError.
Here's a snippet:

Traceback (most recent call last):
  File "/home/john/Dropbox/git/pyoptsparse/postprocessing/OptView.py", line 1606, in <module>
    disp = Display(histList, outputDir, figsize)
  File "/home/john/Dropbox/git/pyoptsparse/postprocessing/OptView.py", line 106, in __init__
    self.OptimizationHistory()
  File "/home/john/Dropbox/git/pyoptsparse/postprocessing/OptView.py", line 232, in OptimizationHistory
    self.DetermineMajorIterations(db, OpenMDAO=OpenMDAO)
  File "/home/john/Dropbox/git/pyoptsparse/postprocessing/OptView.py", line 274, in DetermineMajorIterations
    if i == 0 or db[key]['isMajor']:
KeyError: 'isMajor'

Need to add some logic to avoid this issue.

[johnjasa]

We're waiting to merge this in until we have a discussion on the procedure for PyOptSparse updates going forward.

[eirikurj]

For consistency of the code use numpy instead of the np

[eirikurj]

Instead of using 1.0 use numpy.float_ or numpy.float64