oliverguhr/german-sentiment-bert · Class Probabilities as Output

Aug 12, 2022

Hello everyone,

I wondered if it is possible to receive the probabilities for the different classes as an output when using your package instead of the predicted class with the highest probability?

Thanks in advance for any responses!

tanjafranziska

Sep 6, 2022

Take the original code from https://github.com/oliverguhr/german-sentiment-lib/blob/master/germansentiment/sentimentmodel.py and change the last two lines of the predict_sentiment function to

    probs = torch.nn.functional.softmax(logits[0], dim=-1).tolist()
    label_ids = torch.argmax(logits[0], axis=1)
    return ([self.model.config.id2label[label_id.item()] for label_id in label_ids], probs)

You will get the probabilities for positive, negative and neutral when calling the predict_sentiment function like

    preds, probs = model.predict_sentiment(texts)

oliverguhr

Owner Oct 10, 2022

Thanks @Arcticweasel and @tanjafranziska ! I added an API feature for the class probabilities:

from germansentiment import SentimentModel

model = SentimentModel()

classes, probabilities = model.predict_sentiment(["das ist super"], output_probabilities = True) 
print(classes, probabilities)

['positive'] [[['positive', 0.9761366844177246], ['negative', 0.023540444672107697], ['neutral', 0.00032294404809363186]]]

oliverguhr changed discussion status to closed Oct 10, 2022