Conversation
|
Here, instead of a new parameter, can we make some tests (such as initialize and see if it fails, or use some routine from codecarbon) and automatically determine (something like try and catch)? |
|
Online is always better, or no? |
|
My major concern with online mode is that in some private infrastructures, networked table lookups are unreliable. Specifically in my case, I'm using private infra through the University of Washington. I've seen that when online mode fails, CodeCarbon defaults to Kansas in the middle of USA; Kansas has different emissions from Washington. I would say online mode is always preferable, except for the default after failure that I'm seeing in my case. Others with similar setups would also run into the same inaccurate emissions tracking. In my local environment I tried to use offline mode parameters when configuring |
bruAristimunha
left a comment
bruAristimunha
left a comment
There was a problem hiding this comment.
Small points and it is good for me
|
Suuuper good, many thanks @davisethan |
Signed-off-by: Bru <[email protected]>
…zed tests - Fixed bug where codecarbon_offline attribute was not set when codecarbon_config is None - Added self.codecarbon_offline = False for default configuration - Created comprehensive test suite with pytest parametrization - Tests cover all offline parameters, online mode, and attribute existence - Reduced code duplication by using @pytest.mark.parametrize - All 15 test cases pass (12 run, 3 skipped when codecarbon not installed) Fixes issues in: - examples/tutorials/tutorial_3_benchmarking_multiple_pipelines.py - examples/advanced_examples/plot_filterbank_csp_vs_csp.py - examples/advanced_examples/plot_hinss2021_classification.py - examples/advanced_examples/plot_use_an_X_y_dataset.py - examples/advanced_examples/plot_statistical_analysis.py
|
Thanks for being an active maintainer @bruAristimunha |
2 participants
Described by #955.