Fixing log issue for Label Model #1652
Conversation
|
@bhancock8, I have made the discussed changes. But, the pipeline is failing at the build stage for python3.6, as the tox command does not complete in 15 minutes Travis wait time. Can you please suggest a way forward for this? Thank you. |
| @@ -622,6 +622,8 @@ def _execute_logging(self, loss: torch.Tensor) -> Metrics: | |||
|
|
|||
| def _set_logger(self) -> None: | |||
| self.logger = Logger(self.train_config.log_freq) | |||
| if self.config.verbose: | |||
| logging.basicConfig(level=logging.INFO) | |||
There was a problem hiding this comment.
This appears to be changing the default logging the logging module, not for this specific class's logger. Can you modify just the logging level of self.logger instead?
There was a problem hiding this comment.
I did consider this. But, the logging has been implemented throughout the code at the root level. The logging inside fit() of LabelModel and log() of Logger are using module level logging function logging.info(). Therefore all the logs are generated by root level logger and the only way to set log level for LabelModel specifically is to change the CustomeLogger significantly. However, I don't think that is necessary. Even if I change the root logger level to INFO, it would not interfere with the logging flow. Because, before creating an INFO level log, there will always be a check for the verbose parameter.
There was a problem hiding this comment.
I see. Alright, this looks good to me then!
|
Thanks for the PR! We'll take a look into why the Python 3.6 unit tests are getting stuck, as it appears to be unrelated to this PR. |
|
Thanks @bhancock8, please merge the PR as well, because I don't have write access. |
|
Yep! I'll make sure this gets merged once the unit test build issue is resolved. Thanks @anerirana! |
Description of proposed changes
Changing log level to
INFOwhen verbose is set toTrueinLabelModel.fit()Related issue(s)
Fixes #1638
Test plan
No new test cases are required
Checklist
Need help on these? Just ask!
tox -e complexand/ortox -e sparkif appropriate.