The output of revscoring models on Lift Wing is different from the ORES ones, and this may end up being a problem for people migrating from one tool to the other one. For example, ad-hoc code has probably been written to get the score value from the following ORES response:
{
"enwiki": {
"models": {
"goodfaith": {
"version": "0.5.1"
}
},
"scores": {
"123456": {
"goodfaith": {
"score": {
"prediction": true,
"probability": {
"false": 0.03387957196040836,
"true": 0.9661204280395916
}
}
}
}
}
}
}Meanwhile this is what we return from LiftWing:
{"predictions": {"prediction": true, "probability": {"false": 0.03387957196040836, "true": 0.9661204280395916}}}I would personally vote to change a little our model.py servers to return the ORES response, so that migrating tools to Lift Wing will end up easier. If the output is the same then we'll only need to change the endpoints.