I would like to scan a dataset with a model and collect the output (say, 10 class probabilities for instance). After scanning, I want to know which output corresponds to which data point. The best way I can think of is to make my
DatasetMaker class) to return
(index, data, label) (not just
(data, label)). This way, it doesn’t matter if my datasetLoader shuffles or not (right?). Then, I’ll collect model outputs in a dataframe.
Is there a better way to scan and trace datasets?