Debugging issues with experiments/baoab_vs_vvvr.py

[jchodera]

@maxentile : I'm trying to dig into what might be going on with negative estimates of $\Delta F_{neq}$ from the baoab_vs_vvvr.py example you pointed us to, but after installing the integrator-benchmark package via

python setup.py develop

there seems to be some issue with importing benchmark.testsystems:

lski1962:experiments choderaj$ python baoab_vs_vvvr.py 
Traceback (most recent call last):
  File "baoab_vs_vvvr.py", line 1, in <module>
    import benchmark.testsystems
  File "/Users/choderaj/github/choderalab/integrator-benchmark/integrator-benchmark/benchmark/experiments/benchmark.py", line 5, in <module>
    from code.integrators import LangevinSplittingIntegrator
ImportError: No module named 'code.integrators'; 'code' is not a package

My guess is

• There is some code that hasn't been checked in. master was last updated 24 days ago (!!!)
• You might have some old code installed locally as code.integrators that makes this work locally for you
• You are using a branch other than master. Could this be refactor (updated 19 days ago)?

Any idea what the issue might be?

[jchodera]

The same problem persists if I use the refactor branch:

lski1962:experiments choderaj$ python baoab_vs_vvvr.py 
Traceback (most recent call last):
  File "baoab_vs_vvvr.py", line 1, in <module>
    import benchmark.testsystems
  File "/Users/choderaj/github/choderalab/integrator-benchmark/integrator-benchmark/benchmark/experiments/benchmark.py", line 5, in <module>
    from code.integrators import LangevinSplittingIntegrator
ImportError: No module named 'code.integrators'; 'code' is not a package

[jchodera]

I realize it's also possible that the baoab_vs_vvvr.py example you pointed us to was run before the refactor.

[jchodera]

Aha! I think this is due to the use of relative imports, which only work in python 2.x; we need to switch to absolute imports for python 3.x.

The big question is whether I should do this for the master or refactor branch.

[jchodera]

Whoops, that is just one problem, it seems. I still can't find a code module.

[jchodera]

Hm, this is proving tricky since benchmark is both the name of the package and a submodule inside of benchmark.experiments.

[jchodera]

I think I've fixed this in #3.