can't use ranking with label value >= 31

[alexeib]

gbm = lgb.LGBMRanker(n_estimators=10)
feats = []
labels = []
for i in range(50):
    feats.append([random.random(), random.random()])
    labels.append(0 if random.random() > 0.5 else 31)
gbm.fit(X=feats, y=labels, group=[50])

will get an error:
lightgbm.basic.LightGBMError: b'label (31) excel the max range 31'

using any value less than 31 in labels works fine. using any value >= 31 will always produce the error above, with the "max range" being the max value of the label (but it will always say (31) for label value)

[guolinke]

you should set “label_gain”

[alexeib]

that doesnt seem to be one of the parameters either on LGBMRanker or on the fit() function in sklearn interface?

also, how come it works for labels with values < 31, and why is the message so strange? i looked in the code and can't figure out how it can even enter into the branch with that statement, given the values it is printing out

setting label_gain to any value does change the message, given the code above, it is always:
b'label (1) excel the max range 50
for various values of label_gain

edit: ok it works if i set it to an array that is the size of my maximum label size. thanks for help and would be nice to document!

[alexeib]

closing although i think error message, documentation and sklearn interface could use a touch up :) thanks again

[guolinke]

@StrikerRUS can you help to add this to the document ?

[StrikerRUS]

@guolinke Sure, will do it today.

[Edmondguo]

how to change label_gain in Python API? I can't find it whether in LGBMRanker or in the fit() function.

[StrikerRUS]

@Edmondguo Use **kwargs for any parameter which isn't presented in arguments.

[mayritaspring]

how to change label_gain in Python API? I can't find it whether in LGBMRanker or in the fit() function.

hi Edmondguo, do you figure out this problem? because I also cannot find it.....thank you so much

[cmcai0104]

Hi, @alexeib What's "the size of my maximum label size" means?

if my label is: 1,2,3,...,100,1,2,3,...,200,1,2,3,...,1000, how to set my “label_gain”?

[guolinke]

@cmcai0104
just make sure label_gain[label_value] is valid.