How to do batch prediction?

[nishith006 (Nishith Showri)]

Hey so I have a test dataset and if do learn.predict() and solve it is very slow predicting 20000 images like that. How could I do this ? Should I create a dataloader for test data too or is there any other process. It would be helpful if you provide me with some code

[florianl (Florian)]

first create a test dataloder for your dataset you’d like to predict and feed it to learn.get_preds().

see this post for the code:

How to specify test set in DataBlock?

Thank you both @VishnuSubramanian @sut ! I guess my follow-up question would be how do I get metric on the test_dl? I figure the following code would give me predictions for the test data test_files = get_image_files('... path to test data...') test_dl = learner.dls.test_dl(test_files, with_label=True) pred_probas, _, pred_classes = learner.get_preds(dl=test_dl, with_decoded=True) Is there a way to get labels from test_dl? Is there an easy way to get metric (error rate in this case)?

[nishith006 (Nishith Showri)]

Thank you very much.