Description

gluon.contrib.estimator used a global Logger obtained via
logging.getLogger('gluon.contrib.estimator.event_handlers'). This logger used
to be configured every time a gluon.contrib.estimator.LoggingHandler was
created, which is a bug. We can't modify a global Logger instance whenever the
user creates an Estimator and a LoggingHandler.

Instead, this commit separates the LoggingHandler (responsible for logging
metadata during estimator.fit) from the configuration of the Logger.

We expose the Logger as attribute of the Estimator, and configure it to output
to stdout by default. Instructions are given how users can configure the
Estimator.logger to log to a file instead.

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage:
• Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
• Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore)
• Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL)
• Code is well-documented:
• For user-facing API changes, API doc string has been updated.
• For new C++ functions in header files, their functionalities and arguments are documented.
• For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable
• Check the API doc at https://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html
• To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

Comments

• This change is backward incompatible, but is important to make. It only changes experimental APIs. The reason is that the previous API design is inconsistent and not user friendly. It is inconsistent, because a global object is modified whenever a LoggingHandler object is created; this leads to buggy behaviour if the user creates multiple LoggingHandlers, as all of them will share the same underlying Python Logger. Instead, we should use a separate logger for each estimator.
• This change improves usability, by switching the default log stream from sys.stderr to sys.stdout. The following screenshot highlights the need for this.

Compare this to