CAMMA-public / rendezvous

A transformer-inspired neural network for surgical action triplet recognition from laparoscopic videos.
Other
24 stars 8 forks source link

sklearn version #4

Closed gshuangchun closed 2 years ago

gshuangchun commented 2 years ago

Hi, what is the sklearn version? when I test the same trained model, the scikit-learn==1.0.2 is

mAPI | mAPV | mAPT | mAPIV | mAPIT | mAPIVT -- | -- | -- | -- | -- | -- 84.04 | 52.18 | 32.49 | 33.83 | 26.36 | 21.9

while the scikit-learn==1.1.1 is

mAPI | mAPV | mAPT | mAPIV | mAPIT | mAPIVT -- | -- | -- | -- | -- | -- 81.54 | 49.79 | 27.86 | 22.75 | 13.89 | 8.53
nwoyecid commented 2 years ago

Thank you for the information, we will inspect how the latest update of sklearn lib affects the behaviour of ivtmetrics. For the meantime, we encourage all users to stick with sklearn <= 1.02 for comparable results. Meanwhile, we will be rolling out a custom 'no library' implementation of AP metrics. This will replace the sklearn in our next release.

nwoyecid commented 1 year ago

Value issue with sklearn version update is now resolved in the ivtmetrics v0.1.1 Thank you