Default dt is 0

[sawyerbfuller]

Changes the new default to be dt=0 (continuous-time), rather than dt=None (timebase unspecified) for state space, transfer function, and input output systems. This follows matlab convention as discussed in issue #422

• new default is dt=0 in these three modules (edit: setting now is in a single location: defaults['control.default_dt'])
• to make interconnection behavior match what is described in conventions, a new function: lti.common_timebase(dt1, dt2) was introduced. It performs logic to correctly combine timebases of combined systems. The new function now correctly allows discrete-time systems with unspecified sampling time (dt=True) to be combined with systems with specified sampling time (dt!=0, dt!=None, eg dt=0.1).
• new functions _isstaticgain (edit: now is a method is_static_gain) were added to detect upon initialization whether a state space or transfer function is a static gain. If so, dt=None for such systems so they may be combined with either discrete- or continuous-time systems.
• I needed to take dt out of the arguments list in iosys init functions because library-wide defaults are not available when modules are first imported, which is when default values in function declarations are evaluated. My solution was to move the dt keyword in the **kwargs dict.
• fixes to unit tests
• use_legacy_defaults('0.8.3') reverts to old default behavior of dt=None

[sawyerbfuller]

Let me know what you think. If it looks good I will update documentation in modules and in conventions.rst

[coveralls]

Coverage increased (+0.07%) to 84.278% when pulling dc53eb9 on sawyerbfuller:default_dt_0 into d3142ff on python-control:master.

[coveralls]

Coverage increased (+0.08%) to 84.296% when pulling 9aed7cc on sawyerbfuller:default_dt_0 into d3142ff on python-control:master.

[sawyerbfuller]

some remarks:

• most of the unit tests worked as-is with the change, except for ones that implicitly assumed dt=None during the tests. After making that assumption explicit on such systems, pretty much everything worked
• most of the effort came in implementing a test for static gain in StateSpace and TransferFunction classes (maybe one is needed in iosys?)
• and in standardizing the use of lti.common_timebase everywhere in the software when interconnecting systems
• I tested it on a few pages of my own code full of discrete and cont time systems and everything worked unchanged
• all unit tests work for both dt=0 and dt=None as defaults

[murrayrm]

For the defaults, it seems like most people are going use the same default for transfer functions, state space systems, and I/O systems. Right now they have to set three defaults. Does it makes sense to collapse this to just one configuration variable? Or perhaps allow both: a "global" default_dt with individual overrides if you want them.

Also: since this will nominally break code, I propose putting this into v0.9 (versus 0.8.4, which we are close to releasing).

Will try to take a closer look at the code itself in a bit.

[sawyerbfuller]

Hi Richard, this all makes sense and sounds reasonable. There was some reason (I think it was mostly expedience) for why I didn’t just use a a global default dt - I think I had decided that something about the global defaults hadn’t been completely implemented yet, but let me take a look, I agree that would be the right way to do it if possible.

[bnavigator]

I like it!

[sawyerbfuller]

OK made a few suggested changes:

• default dt is now package-wide. change it with set_defaults('control', default_dt=0)
• re-incorporated timebaseEqual with deprecation warning
• improved a few unit tests

[sawyerbfuller]

latest commit creates a new, potentially-useful method, is_static_gain() for StateSpace and TransferFunction systems. refactored their init methods to use it now rather than what it used to be, which was a function internal to the respective module

• during refactor, added the ability to init using a dt keyword, as suggested in issue Problem with DT State-Space Response #419. If dt is received as both a positional and a keyword argument, the user receives a warning and the positional argument is used.

[sawyerbfuller]

I think this should also resolve #399

[bnavigator]

Did someone check to merge your PRs with @samlaf's? They seem to work on overlapping functionality.

[sawyerbfuller]

Did someone check to merge your PRs with @samlaf's? They seem to work on overlapping functionality.

@bnavigator good to check. A brief glance through suggests that they are are basically operating on different aspects of the iosys functionality, and I don't anticipate any merge problems, except for the occasional empty line. The devil is in the details, of course. I don't know git will enough to do a test merge. If any issues crop up you can let me know and I will be happy to take a look at it.

[bnavigator]

I made a test merge in bnavigator#1 and reactivated TravisCI for my fork. Tests look good so far, there were merge conflicts in both merges, because the edits were in lines next to each other. Please check, if I resolved correctly.

Originally posted by @bnavigator in #433 (comment)

I postet the comment in the wrong thread. Meant to post it here, so that @samlaf gets notified too. Alas, another mention.

[bnavigator]

Superseded by #490