I was playing around for some features and while iterating I was getting bored of waiting for tests.
Thus running pytest . --durations 30 --disable-warnings I could get the most time consuming tests.
The code changes keep the same level of coverage but lower the runtime.
I opened a PR in my own fork to quantify the difference and it is quite nice (4m 14s vs last merge here of 7m 37s).
I didn't investigate much further, but since it was a bit relevant for me I wanted to share it.
Love to hear that!
To be fair, the best improvement comes from not generating the splits in the grouped time series cv, but running its "raising warning" method as standalone
Description
I was playing around for some features and while iterating I was getting bored of waiting for tests. Thus running
pytest . --durations 30 --disable-warnings
I could get the most time consuming tests.The code changes keep the same level of coverage but lower the runtime.
I opened a PR in my own fork to quantify the difference and it is quite nice (4m 14s vs last merge here of 7m 37s).
I didn't investigate much further, but since it was a bit relevant for me I wanted to share it.