mxnet-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From kellen sunderland <kellen.sunderl...@gmail.com>
Subject RE: Improving and rationalizing unit tests
Date Mon, 16 Oct 2017 17:31:29 GMT
I think you’ve covered the pros/cons of having determinism in your tests.  It seems like
a potential maintenance effort versus forced code robustness argument to me.  I’d suggest
you have a separate vote on this topic.

For me the main topic that should be focused on is making the CI system fast and stable for
PR builds.  I think the best road to doing this, as previously suggested, is to segment tests
and move those that are long running, fail occasionally, or require external components into
a nightly/weekly test.

I would also propose that:

6) Test fixtures are created to test a subset of functionality, and that we don’t have test
fixtures like test_operator.py that are nearly 5000 lines long, and take 20 minutes to run.
 There’s a few advantages to breaking these tests into smaller files:  

We will have fewer merge conflicts, because fewer people will be editing the same test files
across PRs.  Debugging issues with tests will become easier, as hopefully there will be less
potential side effects between tests (this does happen now). We may be a little more confident
that the tests run independently, eventually meaning that we could run them in parallel more
easily, which would reduce test run latency time (but not throughput).  Last, we will be able
to disable tests at some convenient level of granularity, for example when running on IoT
devices, or without OpenCV.  At the moment we’d have to disable tests individually.

7) We cleanup tests that are no longer needed:

I’ve personally found it quite unintuitive in MXNet to discover which tests are actually
needed, where they are run, how often, etc.  Are the nightly tests actually being run nightly?
 Are the cpp tests run, why is the Travis CI folder still there, what is the difference between
ci_build folder and the Jenkins folder, etc.  If we’re going to take a look at revamping
the tests folder I’d recommend we clean up the folder structure a bit, and delete the non-relevant
files to make it easier for newcomers to know what’s happening.  We’ll always have these
files for reference in source control.

-Kellen

From: Chris Olivier
Sent: Monday, October 16, 2017 6:46 PM
To: dev@mxnet.incubator.apache.org
Subject: Re: Improving and rationalizing unit tests

My argument is that I am actually categorically against having a
requirement that the same input values be used for testing for every run.

I don't personally view "convenience in reproducing" as outweighing
"finding edge cases that I didn't think of or that haven't been tried
before".

On Mon, Oct 16, 2017 at 9:34 AM, Pedro Larroy <pedro.larroy.lists@gmail.com>
wrote:

> It's always going to be deterministic one way or another unless you use
> random from the entropy pool such as /dev/random. I don't think it's a good
> practice not to seed properly and have values depend on execution order /
> parallelism / time or whatever, but that's just my opinion. I would want to
> use the same values for all test runs for reproducibility.
>
> I think your argument goes more towards the previously mentioned "property
> based testing" approach, which is in the spirit of what you are supporting,
> if I'm not mistaken.
>
> On Mon, Oct 16, 2017 at 6:22 PM, Chris Olivier <cjolivier01@gmail.com>
> wrote:
>
> > My take on the suggestion of purely deterministic inputs is (including
> > deterministic seeding):
> >
> > "I want the same values to be used for all test runs because it is
> > inconvenient when a unit test fails for some edge cases.  I prefer that
> > unforseen edge case failures occur in the field and not during testing".
> >
> > Is this the motivation?  Seems strange to me.
> >
> >
> > On Mon, Oct 16, 2017 at 9:09 AM, Pedro Larroy <
> > pedro.larroy.lists@gmail.com>
> > wrote:
> >
> > > I think using a properly seeded and initialized (pseudo)random is
> > actually
> > > beneficial (and deterministic), handpicked examples are usually too
> > > simplistic and miss corner cases.
> > >
> > > Better yet is to use property based testing, which will pick corner
> cases
> > > and do fuzzing automatically to check with high degree of confidence
> > that a
> > > testing condition holds.
> > >
> > > Probably it would be good if we use a property based testing library in
> > > adition to nose to check invariants.
> > >
> > > A quick googling yields this one for python for example:
> > > https://hypothesis.readthedocs.io/en/latest/quickstart.html does
> anyone
> > > have experience or can recommend a nice property based testing library
> > for
> > > python?
> > >
> > >
> > > Regards
> > >
> > > On Mon, Oct 16, 2017 at 4:56 PM, Bhavin Thaker <bhavinthaker@gmail.com
> >
> > > wrote:
> > >
> > > > I agree with Pedro.
> > > >
> > > > Based on various observations on unit test failures, I would like to
> > > > propose a few guidelines to follow for the unit tests. Even though I
> > use
> > > > the word, “must” for my humble opinions below, please feel free to
> > > suggest
> > > > alternatives or modifications to these guidelines:
> > > >
> > > > 1) 1a) Each unit test must have a run time budget <= X minutes. Say,
> X
> > =
> > > 2
> > > > minutes max.
> > > > 1b) The total run time budget for all unit tests <= Y minutes. Say,
> Y =
> > > 60
> > > > minutes max.
> > > >
> > > > 2) All Unit tests must have deterministic (not Stochastic) behavior.
> > That
> > > > is, instead of using the random() function to test a range of input
> > > values,
> > > > each input test value must be carefully hand-picked to represent the
> > > > commonly used input scenarios. The correct place to stochastically
> test
> > > > random input values is to have continuously running nightly tests and
> > NOT
> > > > the sanity/smoke/unit tests for each PR.
> > > >
> > > > 3) All Unit tests must be as much self-contained and independent of
> > > > external components as possible. For example, datasets required for
> the
> > > > unit test must NOT be present on external website which, if
> > unreachable,
> > > > can cause test run failures. Instead, all datasets must be available
> > > > locally.
> > > >
> > > > 4) It is impossible to test everything in unit tests and so only
> common
> > > > use-cases and code-paths must be tested in unit-tests. Less common
> > > > scenarios like integration with 3rd party products must be tested in
> > > > nightly/weekly tests.
> > > >
> > > > 5) A unit test must NOT be disabled on a failure unless proven to
> > exhibit
> > > > unreliable behavior. The burden-of-proof for a test failure must be
> on
> > > the
> > > > PR submitter and the PR must NOT be merged without a opening a new
> > github
> > > > issue explaining the problem. If the unit test is disabled for some
> > > reason,
> > > > then the unit test must NOT be removed from the unit tests list;
> > instead
> > > > the unit test must be modified to add the following lines at the
> start
> > of
> > > > the test:
> > > >     Print(“Unit Test DISABLED; see GitHub issue: NNNN”)
> > > >     Exit(0)
> > > >
> > > > Please suggest modifications to the above proposal such that we can
> > make
> > > > the unit tests framework to be the rock-solid foundation for the
> active
> > > > development of Apache MXNet (Incubating).
> > > >
> > > > Regards,
> > > > Bhavin Thaker.
> > > >
> > > >
> > > > On Mon, Oct 16, 2017 at 5:56 AM Pedro Larroy <
> > > pedro.larroy.lists@gmail.com
> > > > >
> > > > wrote:
> > > >
> > > > > Hi
> > > > >
> > > > > Some of the unit tests are extremely costly in terms of memory and
> > > > compute.
> > > > >
> > > > > As an example in the gluon tests we are loading all the datasets.
> > > > >
> > > > > test_gluon_data.test_datasets
> > > > >
> > > > > Also running huge networks like resnets in test_gluon_model_zoo.
> > > > >
> > > > > This is ridiculously slow, and straight impossible on some
> embedded /
> > > > > memory constrained devices, and anyway is making tests run for
> longer
> > > > than
> > > > > needed.
> > > > >
> > > > > Unit tests should be small, self contained, if possible pure
> > (avoiding
> > > > this
> > > > > kind of dataset IO if possible).
> > > > >
> > > > > I think it would be better to split them in real unit tests and
> > > extended
> > > > > integration test suites that do more intensive computation. This
> > would
> > > > also
> > > > > help with the feedback time with PRs and CI infrastructure.
> > > > >
> > > > >
> > > > > Thoughts?
> > > > >
> > > >
> > >
> >
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message