The goal of the skops persistence format is not to be fast. Nevertheless, it would be useful for us to measure how fast it is and compare it to pickle. This way, we cannot only observe if there are cases were skops is unreasonable slow, making it less useful in practice. We might also find performance regressions, which could indicate bad implementation or bugs.
Probably, it's overkill to have these tests as part of our regular test suite. But we could think of an automatic GH action that runs on a regular basis and errors if skops performance exceeds pickle performance by a certain margin.
The goal of the skops persistence format is not to be fast. Nevertheless, it would be useful for us to measure how fast it is and compare it to pickle. This way, we cannot only observe if there are cases were skops is unreasonable slow, making it less useful in practice. We might also find performance regressions, which could indicate bad implementation or bugs.
Probably, it's overkill to have these tests as part of our regular test suite. But we could think of an automatic GH action that runs on a regular basis and errors if skops performance exceeds pickle performance by a certain margin.