update analysis.base.AnalysisBase to conform to API

[orbeckst]

In #719 we decided on the details of the analysis interface ("Bauhaus" style). The description for implementation is on the wiki MDAnalysis.analysis user interface but we also decided to change a few additional things that are directly defined in the base class and automatically apply to all analysis code: The following needs to be changed in AnalysisBase:

1. start, stop, step are kwargs of AnalysisBase.run() but not of __init__().
2. verbose is a kwarg of __init__ where it sets the default; verbose is also a kwarg of run() where it overrides the default value for the duration of run (this allows the user to mute the frame-by-frame calculation but still have verbose output during pre- and post-processing).

Because 1 changes the API, we need to deprecate start, stop, step in __init__ and remove them in 1.0.0. Until then it needs to be possible to set the iteration range in __init__ and kwargs in run() will override the __init__ values.

[orbeckst]

@kain88-de can you have a look and see how much work it is to get this done for 0.17.0?

[kain88-de]

From my memory. A problem is that we often use information from the start stop step arguments in the init function. This means all sub classes have to be checked after we add a deprecation to the base class. Verbosity is likely easier. There might be a corner case in the PCA analysis. I’m not sure if there are some surprises in the decorators like base_analysis. With writing tests that’s at least a couple of hours of work.

…

On 27. Oct 2017, at 21:07, Oliver Beckstein ***@***.***> wrote: @kain88-de can you have a look and see how much work it is to get this done for 0.17.0? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

[orbeckst]

@richardjgowers can we get deprecations into 0.19.0 #1971 ?

In particular, the start/stop/step in run() is reversing a decision that we made much earlier when we tried to move them from run to __init__. We have depreciation code that says "start/stop/step will be removed from run() in 0.17.0" :-p.

[richardjgowers]

@orbeckst I don't understand what needs doing for this, do some classes have start/stop/step in run?

[orbeckst]

Classes should have start/stop/step in run() but not in __init__.

At least one (rms.RMSD.run()) has it already in run() but it is deprecated. It should be un-deprecated because it will become the norm.

[jbarnoud]

Here are the things we need to do for this to come true:

• Add start, stop, step, and verbose arguments to AnalysisBase.run.
• Remove deprecated quiet from AnalysisBase.__init__.
• Add deprecation warnings to AnalysisBase.__init__.
• In analysis classes, move some logic from __init__ to _prepare.

[orbeckst]

@nawtrey added a parallel DensityAnalysis in MDAnalysis/pmda#93 which can be back-ported.

[IAlibay]

@orbeckst the removal of start, stop, and step from AnalysisBase.__init__() are next up on my list of things to do for #2443

As it is a bit unclear from the thread, I thought it might be worth asking if there is still work to be done here, or is it ready for removal?

[orbeckst]

• I assume for anything using AnalysisBase, just removing will be fine. Just check the actual code of the analysis classes for any workarounds, deprecation messages, etc.
• Any of the analysis classes that do not use AnalysisBase: keep them as they are as they do not fall under this issue. (If we can get any other classes more in line with the AnalysisBase API for 1.0 then that would be great — but raise separate issues and reference this issue.)

[IAlibay]

@orbeckst I think this was completed with #2505, is this good to close?

[orbeckst]

Yes, thanks for the reminder. This was done with #2505 and the changes to verbose (as part of one of the bigger gsoc PRs (EDIT: the tqdm/progressmeter one)).