What is the difference between test, val and train set

[krzysztofMajchrzak-GIT]

During learning process, the query set must contain images from support set. So we use only training set for building our tasks, I mean query and supports taks. So what is the purpose of test and validation json files? How does algorithm use them?

[krzysztofMajchrzak-GIT]

I am trying to train the network so that it can recognize particular image. If I have omniglot dataset there are many classes in background_images.zip archive. This archive is used to train the network. But if i say that I want to make it 5 way 1 shot, what 5 classes from vinyals_train_labels.json are taken? If I train the network with 5 way 1 shot approach the network should be able to recognize one of 5 particular classes, but which 5? How can i determine what classes the classficator learns to recognize? And if I put the image through already trained model I get 5 scores, from which I take the highest one. But I don't know if I am testing it against the class that the network is aware of. I dont get what is the purpose of vinyals_test_labels.json and vinyals_val_labels.json. One is used for metatesting i suppose, but don't know which one. Very sorry for so many problems but i really need to understand it ASAP. Thank You for Your time in advance.