WIP: Issue 1323 optest#1334
Conversation
|
Bump. Would be nice to get some feedback. |
| # Copyright 2014-2017 The ODL contributors | ||
| # coding: utf-8 | ||
|
|
||
| # Copyright 2014-2017 The ODL contributors |
| """Print message if ``self.verbose == True``.""" | ||
| if self.verbose: | ||
| print(message) | ||
| VERBOSITY_LEVELS = {'DEBUG': 0, 'INFO': 1, 'CHECK': 2, 'WARNING': 3, |
There was a problem hiding this comment.
These are not exported. Is that on purpose? Also, shouldn't these be the same for all the things in diagnostics? Move to a unified file.
| """Estimate the operator norm of the operator. | ||
|
|
||
| The norm is estimated by calculating | ||
| #TODO: add `logfile` parameter |
| logger=self.log) as counter: | ||
| log('', level='SECTION') | ||
| log('Linearity', level='SECTION') | ||
| log('=' * log_linewidth, level='SECTION') |
There was a problem hiding this comment.
Stuff like this has to be solvable in a much nicer way. Perhaps create a log object with some standard calls (e.g. log.new_section('Linearity'), that would make this so much more readable
|
|
||
| # --- Basic properties --- # | ||
|
|
||
| num_failed += _check_cond( |
There was a problem hiding this comment.
Perhaps we could have a utility class for this that contains all executed tests, e.g.
test_history = TestHistory()
test_history.append(_check_cond(...))
print(test_history.num_failed) # 1
|
|
||
| log('', level='SECTION') | ||
| log('Basic properties', level='SECTION') | ||
| log('=' * log_linewidth, level='SECTION') |
There was a problem hiding this comment.
These 3 line logs should be replacable by something standardized to save lots of space and copy-paste code
| kwargs={'verbosity': verbosity, 'deriv_arg': deriv_arg}, | ||
| verbosity=verbosity) | ||
|
|
||
| log('## Getting operator properties...', level='DEBUG') |
There was a problem hiding this comment.
How about using a format like "with (something)" that indents the strings inside it? E.g.
## Getting operator properties...
Got derivative
## Done
| level=failed_level) | ||
|
|
||
| return dict(num_failed=num_failed, | ||
| deriv=deriv, inverse=inverse, adjoint=adjoint) |
|
|
||
| denom = opnorm * (x.norm() + y.norm()) | ||
| if denom == 0: | ||
| denom = 1.0 |
| Wikipedia article on `Adjoint | ||
| <https://en.wikipedia.org/wiki/Adjoint>`_. | ||
| """ | ||
| opnorm = op.norm(estimate=False) |
There was a problem hiding this comment.
There is a utility that does this (in this file)
2 participants
This is an attempt at improving our
diagnosticsmodule. It's work in progress, and I'm curious what others think about the output and the workflow.Here's what I think is better:
OpTestclass is gone, instead everything is implemented with plain functions.norm()being executed before, and it would also fail for nonlinear operators.'DEBUG'which prints A LOT, to'QUIET', which only prints the number of failures.Here's an example:
Output is here
TODO:
{'num_failed': <number of failed tests>}. Maybe some more details would be nice, like number of tests run, elapsed time, or even a list of checks, each with timing and result.Input on this is appreciated!
Updated TODO: try all of the above and see how it goes
examplesin all spaces where definedFrom review:
Loggerclass (or functions) for verbosity level specific logging, to make the code cleanerTestHistoryfor convenient storage (and output) of test results_get_...)Ping @adler-j @aringh @sbanert
Related issues: #1323, #526, #717, #340, #339