Can we overfit batches if labels are incorrect?

[talhaanwarch]

I want to know can we use overfit batches to check the sanity of my labels data? For example if i randomly define the labels or if i shuffle the labels with out shuffling corresponding features, can we still overfit_batches and get 100% results?

[rohitgr7]

up to you, but you shouldn't use incorrect labels with overfit_batches, since the idea here is to check whether the model/data pipeline is good enough for further training or not. If you pass in in-correct labels, then the results won't signify what we are trying to infer.