Hello,
I am working multi-class. When I study at Keras I can use “predict_proba” function for can see probability of every class.
Model is Sequential() and I used CNN.
code: y_score = model.predict_proba(testX)
I want to learn, is there a function at Pytorch like “predict_proba” .
Assuming you are working on a multi-class classification use case, you can pass the input to the model directly and check the logits, calculate the probabilities, or the predictions:
model.eval()
logits = model(data)
probs = F.softmax(logits, dim=1) # assuming logits has the shape [batch_size, nb_classes]
preds = torch.argmax(logits, dim=1)
roc_curve expects the scores as an input, not the predictions, since the curve itself will be created using different thresholds.
From the docs:
y_score ndarray of shape (n_samples,)
Target scores, can either be probability estimates of the positive class, confidence values, or non-thresholded measure of decisions (as returned by “decision_function” on some classifiers).
Target scores, can either be probability estimates of the positive class, confidence values, or non-thresholded measure of decisions (as returned by “decision_function” on some classifiers).
From Pytorch what output i can expect as scores? i think torch.sigmoid return probabilities?
Hi @ptrblck I have a multi class classification setup and I want to deal with out of distribution detection, my model gives unrealistically high confidence for out of distribution as well. I am fairly new in this and I was hoping if you could help me with this? thanks