LightGBM binary classification model: predicted score to class probability

Question

I'm training a LGBM model on a classification (binary) dataset.

import lightgbm as lgb
def lgb_train(train_set, features, train_label_col, sample_weight_col=None, hyp = hyp):
    train_data = lgb.Dataset(data=train_set[features], label=train_set[train_label_col],)
    model = lgb.train(
        train_set=train_data,
        params=hyp,
        num_boost_round=hyp['num_boost_round'],
    )        
    return model

preds = np.array(model.predict(test_features))

Now, the problem is: when I call the predict function, I get a score [0.00012, 0.0035, 0.0000048], how can I calculate the probabilities of each class?

How many samples do you have in your `test_features`? Is it 3? If so, those will be the probabilities of 0 for those data points in your `test_features`. — TYZ, Oct 05 '22 at 22:09

score 0 · Answer 1 · answered Oct 06 '22 at 13:29

0

As you mentioned in the comment section, you have 3 samples in your test_features and you got 3 scores from model.predict, those will be the probabilities for each sample.

Ref: https://lightgbm.readthedocs.io/en/latest/pythonapi/lightgbm.LGBMClassifier.html#lightgbm.LGBMClassifier.predict

answered Oct 06 '22 at 13:29

TYZ

8,466
5
29
60

I don't see where it says that it's predicting the probability of class 0, in the documentation, it just says score / predicted values. Can you clarify this? – Zabir Al Nazi Oct 06 '22 at 13:49
and note, I don't use LGBM classifier. So, I don't have predict_proba method. – Zabir Al Nazi Oct 06 '22 at 14:11

score 0 · Answer 2 · answered Oct 10 '22 at 10:43

0

The model predicts the probability of class 1.

Ref: https://github.com/microsoft/LightGBM/issues/5352

answered Oct 10 '22 at 10:43

Zabir Al Nazi

10,298
4
33
60

LightGBM binary classification model: predicted score to class probability

2 Answers2